-I
Understanding Bus Service Reliability:
A Practical Framework Using A VL/APC Data
By
Laura Cecilia Cham
Bachelor of Science in Civil Engineering
Georgia Institute of Technology, 2003
Submitted to the Department of Civil and Environmental Engineering
in partial fulfillment of the requirements for the degree of
MASTER OF SCIENCE IN TRANSPORTATION
MASSACHUSETT NST lV
OF TECHNOLOGY
at the
Massachusetts Institute of Technology
AR 09 2006
February 2006
LIBRARIES
@ 2006 Massachusetts Institute of Technology
All rights reserved
C)nn)
Au th o r ..............................................
Department of Civil 61d Envig0'nmental Engineering
September 12, 2005
C e rtified by ..................................................
r
Nigel H.M. Wilson
Professor of Civil and Environmental Engineering
Thesis Supervisor
A4 , 11
A
J.....
1nr tevJ. Whittle
Chairman, Departmental Committee for Graduate Students
Accepted by .......................................................
BARKER
Understanding Bus Service Reliability:
A Practical Framework Using AVUAPC Data
BY
LAURA C. CHAM
Submitted to the Department of Civil and Environmental Engineering
in partial fulfillment of the requirements for the Degree of
Master in Science in Transportation
ABSTRACT
Service reliability on a transit system can have significant impacts on its provider and both
existing and potential users. To passengers, unreliable service affects their perception of
service quality and transit utility compared to other mode choices, while to transit agencies, this
translates to loss of ridership and revenues and higher costs to provide additional service to
compensate for poor service operations. The introduction of technologies such as Automatic
Vehicle Location (AVL) and Automatic Passenger Counters (APC) provides the opportunity to
gather large sets of data at relatively low cost and evaluate service to improve performance,
schedule planning and operations control.
This thesis presents a comprehensive review of key elements of service reliability, focused on
the measures of reliability, the causes of unreliability and the application of strategies to improve
service. The most significant causes of service reliability are presented: deviations at terminals,
passenger loads, running times, environmental factors (or externalities) and operator behavior.
Each is reviewed in terms of how they impact service and the complexities and interrelationship
between different causes are explored. Also reviewed are the potential preventive and
corrective strategies, and the links between the causes of service unreliability and best strategy
according to the source of problems.
A practical framework is developed to assess service reliability, exploring the uses of Automated
Data Collection (ADC) systems to characterize service reliability and evaluate the causes of
unreliability that may exist. Its goal is to serve as a guide for transit agencies to begin to
analyze the large sets of data available from these systems to evaluate performance and
implement efficient strategies to improve service planning and operations. The proposed
framework consists of three blocks: 1) characterization of service reliability through service
measures and performance reports; 2) identification of causes of reliability problems; and 3)
selection of strategies which target critical causes of unreliability to improve service.
Characterization of service reliability involves examining five key elements an agency should
analyze: a) data inputs, b) output calculations, c) service measures, d) threshold values, and e)
performance reports. Identifying the causes of unreliability includes two sequential processes to
infer the causes of service reliability problems. The first focuses on deviations at terminals,
because good on-time performance and headway adherence is expected at the terminals and
deviations at this point tend to propagate down the route and create further reliability problems.
The second process examines deviations at other points on the route, and follows a set of steps
to infer the causes of unreliability: initial deviations at terminal, passenger loads, poor schedule
planning, operator behavior and externalities. Application of strategies includes an assessment
of the best strategies to prevent reliability problems and reduce the impacts on service
performance, based on the results of the previous analysis.
The application of the proposed framework on the Silver Line Washington Street in Boston (MA)
revealed that variability of running times and headway distributions are high. This indicates that
bus arrivals and passenger wait times on this route are unpredictable and travel times are
irregular. As a Bus Rapid Transit route, which is suppose to provide bus service with rail transit
quality, headway adherence is poor on this route, with a tendency for buses to bunch together
or leave gaps in service. Further analysis revealed that service reliability has recently
deteriorated as a result of the implementation of a new Automatic Fare Collection (AFC) system.
The new fare collection system presented delays in the boarding process, which resulted in
increased travel times and passenger wait times.
The main cause of service unreliability on this route was identified to be deviations at the
terminals. Trips are departing the terminal with poor headway adherence (and therefore, poor
on-time performance), which propagates and creates further reliability problems down the route.
The causes of these terminal deviations were inferred to be a combination of poor terminal
supervision and operator behavior. Recovery times, externalities and passenger loads at this
terminal are inferred to cause only minor problems. At other points in the route, operator
behavior and passenger loads are observed to affect reliability in the inbound direction.
As for strategies to improve service reliability, emphasis is given to better supervision at the
terminal. Supervisors at terminals are needed to enforce good operator behavior, balance
headways, apply control strategies, and coordinate passenger loads to avoid poor departure
headways and overcrowding of buses. Along the route, operator training, corrective strategies
and traffic signal priority are highlighted as potential strategies to reduce the variability in
running times and balance headways to reduce the occurrence of bunches and gaps in service.
Thesis Supervisor: Nigel H.M. Wilson
Title: Professor of Civil and Environmental Engineering
To My Family
I
ACKNOWLEDGEMENTS
I am grateful to all the people I met during my two years at MIT that taught me valuable lessons,
provided me support and became part of my adventure.
First, I want to thank my advisor, Prof. Nigel Wilson, for his support and guidance all the way
through this research. This thesis would not have been a memorable experience and
challenging journey without his patience, understanding, and critical attention that kept pushing
me to do my best.
I would also like to express my thanks to John Attanucci for his valuable comments and
contributions to this research. I want to thank Fred Salvucci, Mikel Murga and Ken
Kruckemeyer for their wonderful stories and valuable feedback. Many thanks to Ginny Siggia,
for her support and amazing way of always being available and having what I needed.
Thanks to the Massachusetts Bay Transportation Authority for funding this research. Special
thanks to David Barker and David Carney for being available to answer all of my questions and
providing me with the data needed for my case study application.
To all my transportation friends, who not only made classes at MIT bearable, but life more
interesting: Jenn, Jeff, Demian, Bassel, Mike. I want to specially thank James and Dalia, my
"real friends", for helping me keep my sanity with lots of laughs, ice cream and crossword
puzzles. Thanks to KC, Ronak, Dave, and rest of folks at Sidney-Pacific, for being part of my
home away from home. And to Alberto, for being a great friend and a star that brighten my way.
To Tim, for sticking by me through my best (and worse) and always knowing I could do it.
And last, but not least, I want to thank my family. To Suelika, Loraine and Vanessa, who helped
guide me through grad school. To my brothers, Jaime and Jorge, and my sister, Carmen, for
always being a source of inspiration and wisdom. And to my parents, for their unending love,
encouragement and faith that have always given me the strength to believe in myself and do
anything.
TABLE OF CONTENTS
1 . INTR O D U C TIO N TO R ELIA BILITY .........................................
1. 1 M otivation ......................................................................................................................
17
17
1.2 Objectives......................................................................................................................18
1.3 Research Approach ...................................................................................................
19
1.4 O utline of Thesis............................................................................................................20
2 . L ITERA TU R E R EV IEW ......................................................
21
2.1 Reliability: The problem and its effects ......................................................................
21
2.2 The Role of Autom ated Data Collection ....................................................................
2.2.1 Key Dim ensions in Data Collection................................................................
2.2.2 Analysis and Decision Support Tools ...........................................................
2.2.3 Software Developm ent .................................................................................
2.2.4 Findings and G uidelines ...............................................................................
22
23
25
26
27
2.3 Transit Reliability Study .............................................................................................
2.3.1 Traveler Behavior ..........................................................................................
2.3.2 Transit Agency Perspective ...........................................................................
2.3.3 Reliability M easures......................................................................................
2.3.4 Causes of Unreliability ....................................................................................
2.3.5 Strategies to Im prove Reliability ....................................................................
29
29
30
30
32
33
2.4 Portland, O regon Studies ..........................................................................................
2.4.1 Im provem ents in Reliability...........................................................................
2.4.2 Schedule Efficiency and O perator Behavior ..................................................
2.4.3 Headway Deviations and Passenger Loads ..................................................
2.4.4 O perations M anagem ent ...............................................................................
2.4.5 O perations Control........................................................................................
35
35
38
41
42
43
2.5 Research Potential .....................................................................................................
43
3. A FRAMEWORK FOR ANALYZING TRANSIT RELIABILITY.............
45
3.1 From Theory To Practice..........................................................................................
45
3.2 Key Characteristics of Service Reliability .................................................................
46
3.2.1
O n-Tim e Perform ance ...................................................................................
46
3.2.2
Headway Regularity......................................................................................
47
3.2.3
Passenger Loads...........................................................................................
48
3.2.4
Overview of Interactions ...............................................................................
49
9
Table of Contents
49
49
52
54
54
55
56
3.3 Causes of Unreliability ................................................................................................
3.3.1 Schedule Deviations at Term inals ..................................................................
3.3.2 Passenger Loads...........................................................................................
3.3.3 Running tim es................................................................................................
3.3.4 Environm ental Factors..................................................................................
3.3.5 O perator Behavior ........................................................................................
3.3.6 Inter-relationships between Causes .............................................................
3.4 Strategies.......................................................................................................................58
58
3.4.1 Preventive Strategies.....................................................................................
61
3.4.2 Corrective Strategies ......................................................................................
3.5 Relationships between Causes and Strategies .........................................................
65
3.6 Proposed Reliability Analysis Process.......................................................................
3.6.1 Characterizing Service Reliability ..................................................................
3.6.2 Identification of Causes .................................................................................
3.6.3 Application of Strategies................................................................................
70
70
83
94
3.7 Fram ework Sum m ary .................................................................................................
98
APPLICATION TO MBTA ...............................
99
4.1 M BTA Silver Line on W ashington Street ....................................................................
99
4.2 Autom ated Data Collection System s ...........................................................................
4.2.1 Archived Data ..................................................................................................
105
105
4.3 Scope of Analysis and Data Assum ptions ...................................................................
108
4. CASE STUDY:
4.4 Silver
4.4.1
4.4.2
4.4.3
4.4.4
4.4.5
109
Line Reliability Assessm ent ...............................................................................
Running Tim es.................................................................................................109
114
Headway Adherence .......................................................................................
Passenger W ait Tim es.....................................................................................116
117
Changes in Perform ance: M ay 2005 ...............................................................
119
Overall Reliability Assessm ent ........................................................................
4.5 Identifying the Causes of Unreliability..........................................................................120
120
4.5.1 Perform ance at Term inal .................................................................................
125
4.5.2 Causes of Deviations at Term inals ..................................................................
4.5.3 Deviations at Other Points...............................................................................131
4.6 Application of Strategies..............................................................................................134
4.7 Sum m ary of Findings...................................................................................................135
5. CONCLUSIONS ....................................................
10
137
Table of Contents
5.1 Fram ew ork S um m ary ..................................................................................................
137
5 .2 M ajo r F ind ing s .............................................................................................................
13 8
5 .3 F uture Re sea rch ..........................................................................................................
13 9
B IB LIO G R A P H Y ............................................................
141
APPENDix A: RUNNING TIME DISTRIBUTIONS...........................143
APPENDix B: TERMINAL ANALYSIS .......................................
145
11
I
I
LIST OF FIGURES
Figure 3-1. Service Reliability Interactions..............................................................................
49
Figure 3-2. Effects of Late Departure from Terminal .............................................................
50
Figure 3-3. Inter-relationships between Causes of Unreliability............................................
57
Figure 3-4. Operator Behavior Strategies .............................................................................
66
Figure 3-5. Running Time Strategies ....................................................................................
67
Figure 3-6. Deviation from Terminal Strategies ....................................................................
68
Figure 3-7. Passenger Load Strategies ..................................................................................
69
Figure 3-8. Externalities Strategies.........................................................................................69
Figure 3-9. Sequential Process to Identify Causes of Deviations ..........................................
94
Figure 3-10. Application of Strategies....................................................................................
96
Figure 4-1. Silver Line Washington Street Route.....................................................................101
Figure 4-2. Passenger Demand Distribution ............................................................................
104
Figure 4-3. PM Peak Running Time Distribution: Dudley to E. Berkeley - Inbound.................112
Figure 4-4. PM Peak Running Time Distribution: E. Berkeley to Temple - Inbound ................ 112
Figure 4-5. PM Peak Running Time Distribution: Temple to E. Berkeley - Outbound ............. 113
Figure 4-6. PM Peak Running Time Distribution: E. Berkeley to Dudley - Outbound .............. 113
Figure 4-7. PM Peak Headway Ratio Distribution....................................................................115
Figure 4-8. PM Peak Headway Distribution (May 2005)..........................................................118
Figure 4-9. Headway Ratio Distribution at Dudley - Inbound ...................................................
121
Figure 4-10. Headway Ratio Distribution at East Berkeley by Initial Ratio at Dudley .............. 123
Figure 4-11. Headway Ratio Probability at East Berkeley by Initial Ratio at Dudley................123
Figure 4-12. Analysis of Causes of Unreliability at Dudley Square..........................................128
Figure 4-13. Analysis of Causes of Unreliability at East Berkeley - Inbound ........................... 132
13
LIST OF TABLES
Table 2-1. Factors Affecting Reliability .................................................................................
22
Table 2-2. Levels of Spatial and Temporal Detail for Data Capture .......................................
24
Table 2-3. Decision Support Tools and Analyses ..................................................................
26
Table 2-4. System Design and Data Capture Issues..............................................................28
Table 2-5. Recommended Measures of Service Reliability ..................................................
32
Table 2-6. Strategies to Improve Service Reliability .............................................................
34
Table 2-7. Parameter Estimates for Running Time Model.....................................................
36
Table 2-8. Running Time Model.............................................................................................40
Table 2-9. Operator Model Parameter Estimates ..................................................................
41
Table 3-1. On-Time Performance Levels of Service..............................................................
47
Table 3-2. Fixed-Route Headway Adhrence Level of Service ..............................................
48
Table 3-3. Effects of a Late Departure....................................................................................51
Table 3-4. Effects of an Abnormally High Passenger Load ..................................................
53
Table 3-5. Types of Expressing .............................................................................................
64
Table 3-6. Data Inputs for Analysis Process...........................................................................71
Table 3-7. Example of Analysis of the Propagation of Deviations .........................................
86
Table 3-8. Time Point Schedule Deviations vs. Terminal Deviation .......................................
86
Table 3-9. Example of Headway Adherence Analysis ...........................................................
88
Table 3-10. Example of Headway Ratios per Terminal Headway Ratios ..............................
88
Table 3-11. Terminal Departure Schedule Deviation vs. Available Recovery Time................90
Table 3-12. Terminal Departure Deviation vs. Terminal Headway .......................................
Table 4-1. S ilver Line R oute Stops ..........................................................................................
92
102
Table 4-2. Scheduled Departure Times of Sample Trip...........................................................103
Table 4-3. Silver Line Washington Street Route Schedule ......................................................
103
Table 4-4. Silver Line Route Time Point Information ...............................................................
106
Table 4-5. Data Items of Time Point Crossing Record*...........................................................107
15
List of Figures
Table 4-6. S ilver Line T im e Periods.........................................................................................108
Table 4-7. Running Time Analysis: Dudley to E. Berkeley - Inbound ......................................
110
Table 4-8. Running Time Analysis: E. Berkeley to Temple - Inbound .....................................
110
Table 4-9. Running Time Analysis: Temple to E. Berkeley - Outbound ...................................
110
Table 4-10. Running Time Analysis: E. Berkeley to Dudley - Outbound..................................111
Table 4-11. Mean Scheduled and Actual Headways ...............................................................
114
Table 4-12. Coefficients of Variation of Headways ..................................................................
114
Table 4-13. Expected Passenger Wait Times..........................................................................116
Table 4-14. Expected Excess Passenger Wait Times .............................................................
116
Table 4-15. Headway Ratio at Inbound Time Points by Ratio at Dudley Terminal..................122
Table 4-16. Mean Recovery Times at Dudley Square .............................................................
125
Table 4-17. Available Recovery Time vs. Schedule Deviation at Dudley Square....................126
Table 4-18. Poor Performance Operators................................................................................131
Table 4-19. Operator Behavior Analysis ..................................................................................
16
133
1.
INTRODUCTION
TO RELIABILITY
Service reliability significantly impacts transit agencies, their passengers and their
potential users, and therefore is a major concern to transit managers. The degree of
adherence to scheduled times and headways can significantly affect customer
satisfaction and perception of service quality. If buses do not run on schedule, service is
perceived as unreliable, and longer waiting times and crowding will result. To the transit
agency, this translates to a decrease in ridership and revenue, as well as potentially
higher costs for overtime and extra buses to provide enough capacity to avoid some
buses from becoming overcrowded.
This research reviews key elements of service reliability and proposes a framework to
analyze reliability using data from Automated Data Collection (ADC) systems. The
practical framework is aimed to serve as a guide for transit agencies to characterize
performance, identify the causes of service unreliability and select efficient strategies to
improve service quality.
1.1
MOTIVATION
Everyday, transit agencies must deal with questions such as why buses are running late
a large percentage of the time. They need to know what conditions lead to bus bunching
on a route. It is clear that when two buses arrive within seconds of each other, after a
long gap in service, there's a problem with service quality and schedule adherence. It is
this variability and the resulting increase in waiting time, and less the bunched arrival of
buses, that frustrates and annoys customers.
Service unreliability can have great impacts on the system and its users. Unreliable
service increases waiting times for passengers (Wilson et al. 1992) and reduces the
probability of on-time arrival at destinations, which frustrates passengers and decreases
the disutility of transit in traveler mode choice (Abkowitz et al., 1978). It also impacts
passenger loads, as buses bunch and the first bus picks up more passengers and
becomes crowded while the next bus follows with far fewer passengers (Kittleson &
Associates, 2003). Impacts on transit agencies include the increase in costs to provide
additional service to compensate for imbalanced passenger loads (Strathman et al.
2003), and eventually, the loss of passengers and revenue (Strathman et al. 1999).
With the introduction of technologies such as Automatic Vehicle Location (AVL) and
Automatic Passenger Counters (APC), large amounts of data is increasingly becoming
available and it is important to pay attention to how this data can be used to improve
service reliability. Analyzing this data can help characterize reliability problems and help
improve service planning and operations control.
Archived data on vehicle operations and passenger activity is becoming widely available
at relatively low cost (Strathman et al. 2003). The larger data samples allow the transit
17
Chapter 1
agencies to perform statistically valid analyses to monitor performance and improve
scheduling, operations control and traffic engineering (Furth 2000). Automated tools can
be used to provide a better understanding of what creates problems in a system, prevent
such problems through better service planning and operations management, and
develop strategies to correct them once they appear. Automated data collection also
reduces the need for expensive manual data collection, and the various types of
performance reports that can be generated may benefit numerous areas within a transit
agency in providing feedback to improve service (Kimpel et al. 2004).
The list below summarizes some of the benefits of automated data collection systems
compared to traditional data collection practices:
1.2
-
Reduced labor: automated data collection systems reduce the need for human data
collectors (e.g. for manual ride checks and point checks) and for post processing
tasks such as entering the data into a computer for analysis.
-
Lower marginal costs: compared to manual collection, automated data collection
systems have higher initial capital costs and lower marginal costs. For example,
obtaining an extra day of data is less costly with automated systems because the
data is collected continuously, in situations where paying the full cost of an extra day
of manual collection could never be justified.
-
Fewer opportunities for human error: the reduced reliance on manual collection and
processing reduces the chance of human error in collecting and transferring data. Of
course, automated data collection systems can also produce errors, but it should be
easier to identify when the systems need repair or maintenance because of the large
data sets typically available and the use of automated error-checking software
routines.
-
Greater data accuracy: data from automated systems needs to be evaluated for
precision and bias, but higher accuracy is possible because of the large sample
sizes. In addition, they eliminate the problem of false data reported by unmotivated
traffic checkers.
OBJECTIVES
This thesis explores the potential uses of Automated Data Collection (ADC) systems
data to better understand the operation of bus routes and the dynamics of service
reliability. With the availability of larger ADC datasets, there is an opportunity for transit
agencies to answer important questions regarding service reliability, such as:
18
-
What conditions lead to service reliability problems?
-
What strategies can be applied to avoid or reduce poor service quality?
Introduction to Reliability
-
How can data from ADC systems be analyzed to improve service planning and
operations monitoring and management?
The objective of this research is to provide transit agencies with a framework to address
these types of questions. By better understanding the dynamics of transit service, transit
managers would be able to improve service through:
-
Improved service delivery: provide promised level of service, and operate at
scheduled times and headways. Respond to actual passenger demands, service
environments and trouble spots.
-
Improved service management: monitor system performance, including allocating
resources efficiently and evaluating operators.
-
Improved service planning: adjust timetables to reflect realistic vehicle running times
and passenger demand.
This research recognizes that the levels of complexity surrounding service reliability and
the very nature of transit systems do not allow for a clear, simple solution to the
problems of reliability. One of its goals is to provide insight into the dynamics of service
reliability and develop a practical framework through which transit agencies can better
identify the causes of problems and select strategies to overcome them.
1.3
RESEARCH APPROACH
The focus of this research is to develop a framework to measure and analyze service
reliability of bus routes using automated data collection systems. The first step of this
approach involves the evaluation of causes of service unreliability, considering the
impacts of each on the bus route or system and their interactions. The framework
identifies certain trigger events, both operational and external, which can cause vehicles
to deviate from their scheduled times or headways and potentially lead to unreliability.
This part of the approach is aimed at developing a means to describe how many
unreliability problems can be explained by these trigger events and how these triggers
propagate unreliability along the route. The triggers evaluated in this research are: late
departure from terminals, unusual passenger loads, inadequate or too generous running
times, traffic conditions and operator behavior.
The next step is to outline service metrics to characterize reliability and service quality.
The metrics provide an overall picture of service reliability to help develop hypotheses
about the causes of unreliability on a particular route. The metrics are intended to reflect
the perceptions of service quality from the perspective of travelers and transit agencies,
and thus they characterize service in terms of on-time performance and variability.
The framework then presents the potential uses of AVL and APC data in analyzing
reliability and computing the relevant service metrics. The objective is to review
available ADC systems and analyze what these systems can tell us about service
19
Chapter 1
reliability and the effects of problems. The framework includes an outline of how data
from these systems can be processed to calculate service reliability measures related to
running times, and schedule or headway adherence.
The framework also considers strategies to counteract service unreliability with the goal
of improving service quality. Evaluation of strategies is based on data analyses to
identify causes of unreliability, and the strategy's ability to minimize the effects on
passengers and to efficiently return to normal service. The strategies are categorized as
preventive or corrective/restorative. Preventive strategies are aimed at preventing
reliability problems through avoidance of its causes. Examples of preventive strategies
are exclusive bus lanes and traffic signal priority schemes. Corrective or restorative
strategies, such as holding or expressing buses, are implemented when problems have
developed and are meant to restore normal service and avoid the propagation of
problems down the route or day.
A case study of the Massachusetts Bay Transportation Authority (MBTA) Silver Line
Washington Street bus route is presented. Various tools are developed in this process
to use the available AVL data and summarize service reliability measures in order to
evaluate the performance of this Bus Rapid Transit (BRT) line. The case study
illustrates the application of the framework including proposed service measures, the
causes of unreliability and potential strategies.
1.4 OUTLINE OF THESIS
Chapter 2 summarizes and reviews previous work relating to service reliability, the
measures and strategies used in operations control and service planning, and the use of
automated data collection systems for service analysis. Chapter 3 describes the
framework analysis process which is used to evaluate service reliability, identify service
metrics, investigate possible causes, and select strategies. Chapter 4 presents a case
study of the application of this methodology to the MBTA's Washington Street Silver
Line. Chapter 5 completes this thesis with conclusions and suggests future research
directions on service reliability.
20
2.
LITERATURE REVIEW
A review of previous studies reveals that considerable research exists on the effects of
transit service reliability on passengers and transit agencies, on the selection of reliability
measures, on the identification of the causes of unreliability, and on the application of
strategies to improve service. However, the complexities of transit service and the
limitations of data collection have made it difficult to fully understand service reliability
and to identify the relationship between service attributes and reliability. In recent years,
the development and implementation of automated vehicle monitoring and related data
collection systems has sparked the opportunity for more detailed analyses.
A comprehensive study published in 1978 (Abkowitz et al.) provides perhaps the most
detailed examination to date of transit service reliability. It presents a framework for the
evaluation of techniques to improve operations, service management and schedule
planning. This research builds on the findings of that study and goes further taking
advantage of automated data collection systems. The prevalence of these new
technologies has changed the fundamentals of data availability and the prior high costs
of manual data collection.
The remainder of this chapter is organized as follows. The first section (2.1) introduces
service reliability while Section 2.2 presents studies related to the roles Automatic
Vehicle Location (AVL) and Automatic Passenger Counters (APC) systems play in
measuring transit reliability. Section 2.3 summarizes the comprehensive transit reliability
study (Abkowitz et al. 1978). Section 2.4 reviews the results of a number of studies
focused on the application and potential of AVL and APC systems in the Portland, OR
metropolitan area to improve service reliability. Section 2.5 summarizes the literature
and the key findings which research builds upon.
2.1
RELIABILITY: THE PROBLEM AND ITS EFFECTS
Service reliability can be defined in terms of the variability of service attributes and its
effects on traveler behavior and on agency performance (Abkowitz et al. 1978).
Reliability problems are often attributed to the dynamic nature of the operating
environment. (Abkowitz 1983). Providing reliable service means keeping buses on
schedule, maintaining uniform headways and minimizing the variance of maximum
passenger loads (Levinson, 1991).
An added complexity in measuring performance and the effects of improvements in
reliability is the differences in the perceptions of reliability by travelers and operators.
Several studies have explored the effects of transit reliability from the perspective of both
transit customers (Abkowitz et al. 1978; Bates et al. 20011; Prioni and Hensher 20001)
As referenced in Strathman et al. 2003
21
Chapter 2
and operating agencies (Abkowitz et al. 1978; Furth 2000). For frequent service,
travelers tend to focus on headway regularity, on-time arrival at destinations, and wait
time, while agencies often see schedule adherence as an important and easier-to-collect
measure of the effectiveness of service delivery.
Table 2-1 summarizes a number of factors that affect service reliability. More detailed
discussions of some of these factors is presented in Chapter 3.
Table 2-1. Factors AffectingReliability
Description
Factor
Traffic conditions
For on-street, mixed-traffic operations, it includes traffic congestion,
signal delays, parking, incidents, etc.
Creates delays and may force detours
Influences the probability of breakdowns
Involves the availability of vehicles and operators to operate scheduled
trips
Transit preferential treatment
Includes exclusive bus lanes and conditional traffic signal priority
Reflects ability to operate under normal conditions and loads with
Schedule achievability
sufficient recovery times to allow most trips to depart on-time
Describes loads between successive buses and from day-to-day
Evenness of passenger demand
Differences in operator driving skills
Involves route familiarity and schedule adherence (particularly in terms
of early running)
Wheelchair ramp and ramp usage
Includes frequency of deployment and amount of time required
Route length and the number of stops
Relates to the exposure to events that may delay a vehicle
Operations control strategies
Application of actions to counteract reliability problems as they
I develop.
* Source: Transit Capacity and Quality of Service Manual - 2"d Edition
Road construction and track maintenance
Vehicle maintenance quality
Vehicle and staff availability
Negative impacts of service unreliability include additional wait time for passengers
(Wilson et al. 1992), overcrowding and the potential need to provide additional service to
neutralize imbalanced loads due to headway irregularity (Strathman et al. 2003).
2.2
THE ROLE OF AUTOMATED
DATA COLLECTION
Extensive research exists on the potential of automatic vehicle location (AVL) and
automatic passenger counters (APC) systems to improve operations, performance
monitoring, scheduling and planning (Wilson et al. 1992, Furth 2000, Furth et al. 2003,
Wile 2003, Kimpel et al. 2004, Hammerle 2004).
It is generally agreed that automated data collection systems present the opportunity to
do statistically valid analyses on service reliability for the first time (Furth 2000, Kimpel et
al. 2004).
Furth et al. (2003) reviews past and potential applications of automatic data collection
systems in service planning, scheduling, performance evaluation and system
22
Literature Review
management. The study describes a number of analysis and decision support tools that
have been, or could be, developed using the output of these systems.
The first chapter of this study describes the historical uses of AVL and APC systems, the
technological advances in recent years and a number of data capture and matching
issues. Historically, for most AVL systems, the focus has been on its real-time
application for operations control and emergency response. The authors state that
many systems were not designed to provide useful archived data because the transit
agencies did not specify such use during procurement and design. On the other
hand, APC systems were designed for off-line analysis and have been used to evaluate
performance. The limitations of APC systems have been the large costs of
implementation and maintenance, and the need to develop software (mainly in-house) to
analyze the data. Other systems that have been implemented include event recorders,
which have been more popular in Europe, and standard vehicle monitoring devices,
such as odometers, engine heat sensors and fareboxes.
Among the technological advances outlined in the study are:
-
The development and improvement of location-based tools such as global
positioning systems (GPS) and geographic information systems (GIS).
-
The evolution of computers, making them smaller, faster, less expensive and with far
more storage capacities. Communication has also improved with wireless network
technologies and more efficient radio communications.
-
The integration of systems and "smart bus" design, which combines the
functionalities of various systems such as passenger information, fare collection and
scheduling.
The study also outlines common issues of matching captured data with the
corresponding schedule. Matching captured data with bus stop locations has been
difficult because many agencies do not have an accurate "stops" database because it
requires a lot of work to maintain and update. However, the introduction of stop
announcements systems tied to GPS systems has increased the need for more accurate
stop matching because errors are very obvious in real-time announcement systems.
Problems also arise when trying to match data to schedule files for the current timetable.
An automated process is needed which requires compatibility between the different
systems. Matching is also limited by the need for valid entries for operator, run and trip
numbers. Issues also exist with end-of-line data records because passenger counts
may be skewed by confusion surrounding the end of the trip and the beginning of the
next trip.
2.2.1 KEY DIMENSIONS IN DATA COLLECTION
Furth et al. (2003) define four key dimensions in archived data: level of spatial and
temporal detail, complete vs. exception data, fleet penetration and sample size, and data
23
Chapter 2
quality control. The first is a hierarchy, summarized in Table 2-2, on the levels of details
available depending on the type of automated systems deployed.
Table 2-2. Levels of Spatial and Temporal Detail for Data Capture
Level
A
B
Description
Event Records
Event-Independent
Records
AVL without real-time
Infrequent (typically
tracking
60 to 120s)
AVL with real-time
Infrequent (typically
tracking
60 to 120s)
Between-Stop
Performance Data
-
Each time point
-
Each stop
C
APC or event recorder
D
Event recorder with
Each stop and
Recorded events
between-stop summaries
between-stop events
and summaries
All types
All events, full speed
E
-
Event recorder / trip
Very frequent (every
recorder
second)
I
profile
Source: Uses ofArchivedAVL-APC Data to Improve Transit Performance and Management.
(Furth et al. 2003)
*
Level A is representative of older AVL systems, where location data was obtained by
polling for the location of buses at certain time intervals. Level B creates a data record
when the vehicle traverses a timepoint, identified through GPS technology or dead
reckoning, using the odometer. For Level B, the data may be transmitted over the air or
stored on-board for later download. Level C, which is mainly applied for APC and stop
announcement systems (rarely with AVL), creates a record for each stop which can
include the time at stop, passenger counts, wheelchair lift or ramp use, and other special
events. Data is typically downloaded at the end of the day, although newer systems are
able to transmit the information through radio communications. Level D includes events
at stops, just like level C, but also records other activities in-between stops, such as
speed and direction change. Level E is the most complete, recording data about time,
location, events and status in an almost continuous manner. It requires the greatest
storage capacity.
The second dimension deals with whether data records are produced routinely, defined
by a time interval or a location-related event, or it is triggered by an unanticipated time of
an event or an unanticipated even itself. The latter is called exception reporting, in
which any non-standard conditions are recorded. Headway and running time analyses
are limiting with exception data because fewer data records are provided, typically with
larger margins of error, which is why most agencies have moved away from developing
performance analyses using exception data only.
The third dimension is fleet penetration and sample size. It is standard with AVL
systems to have the entire fleet equipped, while it is typical to have only a subset of the
fleet equipped with APC systems. The sample size is related to the fleet penetration and
the rate of sampling. Having only a subset of the fleet equipped with an ADC system
decreases the number of valid headways recovered and considerations in vehicle
assignment become important to ensure route and daily variations are taken into
24
Literature Review
account. With the entire fleet equipped, transit providers benefit from larger sample
sizes to estimate any significant variability as well as the response from any changes in
service or demand.
The last dimension described by Furth et al. is the need for good quality control and
post-processing to correct for any errors. Errors in identifiers and passenger counts are
inevitable, and should be flagged and omitted from any analysis.
The study also states that good analysis using archived data from ADC systems
depends on the availability of other useful data items and databases. The capabilities of
AVL and APC systems are enhanced when other related data items are integrated.
Potentially valuable data items include: door open and close times, start and stop times,
time stamps on passenger entries and exits, off-route events, mechanical and security
alarms, communications to and from the control center, traffic control messages, farebox
transactions, and annunciation and destination signs. Related databases include
schedule data, GIS, payroll, farebox, maintenance, weather or special events, and
customer satisfaction.
2.2.2 ANALYSIS AND DECISION SUPPORT TOOLS
Furth et al. also identify a number of analyses and decision support tools, as shown in
Table 2-3, and used in current transit practice as well as potential functions that would
improve service management and performance. The usage code indicates the level of
use by agencies with AVL-APC data, where [4] indicates the tool is commonly used by
agencies and [0] not used.
25
Chapter 2
Table 2-3. Decision Support Tools and Analyses
Function
General service monitoring.
- Did service operate as schedule?
Targeted investigation
- 'Replay service to review customer
serviceecompiaints, security (incidents,
accidents}. operator Performance
Schedule and Monitoring Running
Time
- Observing running times based on
scheduled running times
Schedule Adherence and Connection
Protection
- Evaluating service and operation
quality
Headway Analysis
- Evaluating service and operation
quality
Demand Analysis
- Analysis of passenger counts
Other Operations Analysis
Too I Analysis [Usage Codej
Missed trips [1]
- Schedule adherence [4]
Trip investigation at gross level (was the trip there? Off-route?) [4]
-
Trip investgation:
early, iatevvercrowded?
[3]
Trip investigation: speed, acceleration [21
Route and segment analysis (mean. distribution) [4]
Suggesting running ime [3] or half cycle [2] based on percentiles
Anaysis of net holding time [2]
Speed and traffic delay [2]
Unsafe operations monitorino [Dl
Relating running time to weather, incidents or special events [11
Percent early, lIate by timepoint [4j
Distribution of schedule deviation at timepoint [3}
Graphical display of schedule deviation distribution along a route [2]
Connection protection [1]
Headway deviations (mean, distribution) [3]
Waiting time impacts (random Passenaer arrivals) [11
Plot successive trajectories (bunchina analysis) 121
Load profie [4] and variations 13
Anaysis of maximum loads and max load point [1]
Time-dependent demand and load analysis [1}
Analyze overload, lift and other events [3]
Transfer and linked trip analysis 1}
Operator performance [1]
Dwell time analysis [2]
Layover and pull in/out analysis [0]
Control effectiveness
Before/alter study, special event or weather analysis
*
Detail needed
A or B
A
C
D or E
B
B
C
0
D or E
B
B (timepoint)
or C (stop)
B or C
B or C
1
B (timepoint)
or C (stop)
C
C
C
C
C
C
C
B (timepoint)
or C (stops)
C
B
By analysis
By analysis
Source: Uses of Archived AVL-APC Data to Improve Transit Performance and
Management. (Furth et al. 2003)
2.2.3 SOFTWARE DEVELOPMENT
The study (Furth et al. 2003) also reviews practice relating to software
development for data analysis for the following five sources of software used to
analyze archived data.
26
Literature Review
-
In-house software. This type of software provides the greatest flexibility because it is
developed for the data needs of the transit provider. Examples of transit agencies
that have developed their own software include King County Metro, Tri-Met and
Metro Transit (Minneapolis/St. Paul). Software developed in-house has the
drawback of requiring personnel and expertise to develop and maintain software, as
well as resulting custom products that cannot be easily transferred to other agencies.
-
Equipment vendors. Software from AVL and APC vendors has been limited to realtime applications (in the case of AVL) and basic report functionalities (for APC
systems). These limit the flexibility and the types of analyses possible because the
needs of the transit provider are not directly tied with generic software. There is also
the risk of poor customer-support from the vendor or of the vendor going out of
business.
-
Scheduling system vendors. Software provided by scheduling system vendors also
have limited flexibility, providing suggested scheduled running times but limited
analysis of schedule or headway adherence. The risks and disadvantages are the
same as software from equipment vendors.
-
Third party software. The advantage of these is that they are not tied to any specific
system, and have more room for improvement and flexibility. Examples of these
include TriTAPT (Trip Time Analysis in Public Transport), developed at Delft
University of Technology's Transportation Engineering Laboratory in the
Netherlands. The disadvantage is that, in the US, funding issues do not allow transit
agencies to purchase these software packages justified only by data analysis. One
way to get around this is to procure these software packages as part of procurement
of the larger AVL or APC systems.
-
Research-Oriented Analyses. These are specialized tools developed by university
research teams. Unless developed in conjunction with the transit agency, like the
partnership between Tri-Met and Portland State University, there is the risk of staff
analyst not being able to apply the specialized software and analyze future data.
2.2.4 FINDINGS AND GUIDELINES
The study develops findings and guidelines in four major areas: system design and data
capture, analysis and decision support tools, quality and integration of other data
sources, and organizational issues. Issues in system design and data capture are
summarized in Table 2-4.
27
Chapter 2
Table 2-4. System Design and Data Capture Issues
Issue
Stop vs. Timepoint
level detail
Time-al-location vs.
Location-at-time
Between-stop
records
Exception data only
Description
Whether data is collected at every sop or only at
timepoints is a key system design decision.
Timepoint level detail is adequate for schedule
planning and adherence analysis, while stop level
detail is better for inlegration with other data items
such as door open/close.
Time-at-location is usually captured through real-time
tracking, while location-at-time is captured by polling
vehicles.
Are full details necessary in general analysis or are
summaries sufficient?
Capturing exception data only is useful in real-time
operations, but very limiting for performance analysis
using archived data.
Findifg and Guidance
Stop level detail also has an advantage on:
- accuracy and end-of-line issues
- wait times, holding time analysis
- posted schedules at stops, support for signal
priority
- better integration with other data systems
The first is preferred because most performance
reports refer to arrival and departure times at
specific points along the route
Full details are helpful for incident investigation.
Summaries sufficient for speed analysis
Exception data cannot be used by itself for
performance reporting, and must be complemented
with other data items.
Central on-board
computer with united
location capability
and interface
The idea of a 'smart bus' design, where location data
is supplied by one central computer. It has
integration capabilities, with a single interface and
shares its location data with other systems.
Accuracy of location inforrnation improves and
matching is easier. A single operator interfaces
reduces the error rate in identifiers.
ID verification: real-
Matching of observed and scheduled data is
time vs. off-line
improved
Real-time ID verification is preferred because
remedial action is taken right away and the nurnber
with valid
sign-in data.
Opportunities to
maxinize valid sign-in data:
- Single interface for operator sign-in
- Range and validity checks during sign-in
- Automatic sign-in, smart-card ID
of records affected is reduced- However, postprocessing corrections can also be automated to
improve data quality.
rinitations on radio bandwidth restrict the amount
On-board data
Radio transmission to central computer is needed for
recording vs. overthe-air transmission
real-time operations control. Archiving done on-board
or by central computer during transmission.
of real-lime APC data that can be transmitted (on
top of the AVL that is already beina sentl.
Data from single-
Use of archived data captured from passenger
purpose systems
Size of equipped fleet
information systemsIt is customary to equip 100% of the fleet with AVL
systems, while only 10%-I 5% with APC systems.
Trend is increasing towards the suppliers offering
off-schedule data and reporting capabilities.
Passenger count analLyses can be done with ony
Data on control
decisions
Supervisors do not usually log any control action
decisions in useful archived data files.
10- 15% of the fleet equipped with APC systems.
(100% equipped fleet if there is a need to report
extreme values). For operations data, it is better to
equip 100% of the fleet.
Need to capture and code any control actions to
flaa records and anabize effectiveness.
Location and farepayment systems
Time-stamping and integration of location records
with fare collection data-
Integration can be done in real-time or off-line, and
benefits include originidestination analysis.
* Source: Uses of ArchivedAVL-APC Data to Improve Transit Performance and Management.
(Furth et al. 2003)
28
Literature Review
In the area of analysis and decision support tools, the study notes the opportunities
presented by the transition to a data-rich environment and its accompanying analysis
possibilities. Service standards, limited previously by the data, can now be revised to
reflect and evaluate the occurrence of extreme values and percentiles instead of just
averages. The study highlights the need for good data integration with related
databases of stop locations, schedule information and fare collection data.
2.3 TRANSIT RELIABILITY STUDY
Abkowitz et al. (1978) present the first comprehensive assessment of transit service
reliability. The study explored the impacts of transit unreliability on both travelers and
transit agencies 2. It presents a framework to develop appropriate measures of reliability,
possible causes of unreliability and effective strategies to overcome the problems. The
main parts of this study are summarized below.
2.3.1 TRAVELER BEHAVIOR
Prior surveys indicated that reliability is ranked by travelers as one of the most important
service attributes. The theoretical framework considers the influence of unreliability on
travel behavior and mode choice, and potential benefits resulting from improvements in
reliability.
The authors review previous studies of the relationship between reliability and travel
behavior, in particular mode choice and departure time decisions. Travelers choose a
travel alternative and departure time based on their perception of travel time variability
and on their risk aversion. This choice reflects the assumptions that travelers associate
a certain loss in value with being late at their destination. They aim to minimize total
travel time, accounting for the variability of in-vehicle travel time and wait time. For the
latter, it was found that travelers do not perceive a reliability problem if bus arrivals are
predictable to some degree. Travelers tended to adjust their arrival time at bus stops to
reduce their expected wait times. The theoretical view of departure time behavior was
based on the travelers' need for on-time arrival at the destination, their familiarity with
the system, and bus arrival patterns.
The authors state that a number of benefits to travelers can accrue from improved
service reliability. The most important is the increase in the probability of on-time arrival,
as the variability in total travel time, especially wait time, is reduced. This probability
increase may decrease the disutility of transit relative to other modes and thus lead to
more frequent use of the system. It may also attract new riders whose transit disutility is
now lower than their prior travel alternative. Reducing travel time variability may allow
travelers to choose a later departure time with the same probability of on-time arrival.
2 The
1978 Transit Reliability Study refers to transit agencies as "Operators". The terminology has been
changed to follow the convention of this research in referring to "operators" as the vehicle drivers.
29
Chapter 2
2.3.2 TRANSIT AGENCY PERSPECTIVE
To transit agencies, reliability is most often defined in terms of schedule adherence.
Prior surveys indicated that service reliability is important to transit agencies in
maintaining efficient operations and providing good service to customers.
A survey of ten bus systems showed the variability in agency behavior regarding key
issues in bus service. On scheduling practices, all ten systems kept a tight scheduled
running time, with no slack time built in, and relied on recovery time to account for the
variability in running times. Even so, only four of them stated that reliability, based on
ensuring on-time departures for successive trips, was a primary factor in determining
recovery times. Four agencies stated that recovery times were mainly set as a result of
union contracts, and two agencies stated recovery times were set to achieve clock-face
schedules. The survey also revealed differences in policies regarding the number of
spare vehicles and drivers. Availability of these spares affects service reliability in terms
of ability to cover vehicle breakdowns and additional service disruptions. All the systems
assigned a certain percentage of their vehicle fleet as spares, but the strategies for
assigning spare drivers varied widely, including a set number per garage, informal
drafting, and daily variations. All agencies identified problems through customer
complaints and monitoring techniques such as road supervisors, ride checks and point
checks. The type, frequency of these monitoring techniques as well as the resulting
actions. Control practices also varied, focusing mainly on recurring problems, and
included adding extra service, rerouting and schedule adjustments.
The theoretical discussion presented in Abkowitz et al. (1978) examines agency
behavior as it affected transit service reliability. Decisions of service operations
managers involve a trade-off in resource allocation and operating policies such as
setting frequency, running times and recovery times. To the agency, reliability strategies
consist of maintaining schedules as reflected through on-time performance measures
and ensuring an adequate availability of spare vehicles and drivers.
The authors state that improving service reliability may yield benefits in reduced capital
and operating costs as a result of reduced travel time variability. Schedule adjustments
to improve service may reduce the size of the spare fleet required. Improving service
reliability also has the potential to increase ridership, and therefore, revenues.
2.3.3 RELIABILITY MEASURES
The study reviews research prior to 1978, highlighting the typical service reliability
measures. The evaluation of these measures revealed several weaknesses at the time
of the study:
-
30
Measures tended to focus on published schedules. As a result, inefficient schedules,
with inadequate running times and recovery times, might have skewed the results of
measures of lateness and deviations.
Literature Review
-
-
A number of the reviewed measures did not capture the effects of reliability problems
from the perspective of travelers.
Variability across different time periods was often overlooked. Data collection
limitations led to a lack of consideration of daily and seasonal variability, and result in
inconsistent comparisons of measures.
The authors developed criteria to address the above weaknesses and proposed a set of
service measures that characterize reliability and are useful to make appropriate
comparisons, help identify problems and select effective remedial strategies. The
reliability measures reflect both the perspectives of travelers and operating agencies,
and focus on:
-
Compactness of distribution to describe the deviations from the average value.
Deviations from the average observed running time, and not printed schedules, are
better measures of variability.
-
Likelihood of extreme delays. To travelers, it measures the probability of long wait
times and late arrival at a destination. For operators, it translates into the likelihood of
a system failure and the need for extra vehicles to restore service due to delays.
-
Normalization of measures. Standardization of indicators allows for comparative
analysis between time periods and routes, and more accurately measure the "true"
impacts of service attributes. For example, standard deviations for routes with very
different headways cannot be usefully compared unless they are normalized by the
average headway value.
The authors review some basic measures of travel time distributions, such as standard
deviation, coefficient of variation, and percent of late trips, to evaluate their advantages
and disadvantages in describing service reliability. Recommendations were against the
use of extreme values for skewed distributions, on ensuring appropriate description of
distributions using the mean and measures of compactness, and on including the mean
value of the attribute for standardization.
Table 2-5 shows the recommended measures based on the arguments. The measures
should account for time-of-day and day-to-day variability, and it was postulated that
analyses ought to consider the interrelationships between different measures and the
development of mathematical relationships to better understand the interactions and
effects of service attributes.
31
Chapter 2
Table 2-5. Recommended Measures of Service Reliability
Distributions of travel time (total
travel, in-vehicle, wait times).
1.
2.
Schedule adherence, measured
at any point along the route.
3.
1.
2.
3.
1.
2.
3.
Distribution of headways
Seat Availability
1.
Mean.
Coefficient of variation (for skewed distributions, standard
deviation should exclude extreme values).
Percent of observations 'N' minutes greater than the mean value.
Average deviation from schedule
Coefficient of variation (from average deviation, not schedules)
Percent of arrivals N minutes later than average deviation from
schedule
Mean.
Coefficient of variation.
Percent of headways:
a. Greater than X percent of average or scheduled
headways, where X 1
b. Lower than Y percent of average or scheduled
headways, where Y 5 1
Passenger loads (demand and capacity)
* Source: Transit ReliabilityStudy. (Abkowitz et al. 1978)
2.3.4 CAUSES OF UNRELIABILITY
The study classifies basic causes of unreliability as environmental or inherent. The
authors consider environmental factors to be those resulting from the surrounding of the
system and are generally random in nature, while inherent factors are those associated
with the transit service itself. The authors separate the causes of unreliability common
to all transit systems, as well as those specific to fixed route bus system.
Common factors for all transit operations included:
32
-
General traffic conditions affect transit service, since transit vehicles typically operate
in mixed-traffic. Variations in travel time result from interactions with other vehicles,
including accidents, turning movements, illegal parking or speed changes.
-
The presence of signalized intersections along the route interrupts the free flow of
traffic and increases the probability of delays. Running times increase due to stopand-go operations and stop times at red lights.
-
Demand varies by day-of-week and season-of-year. Agencies account for
systematic or known fluctuations in demand by adjusting schedules accordingly, but
random variations in demand cause variability in dwell times and travel times.
-
Vehicle and driver availability is related to the operating policies of the agency, but
affects reliability in terms of being able to cover all scheduled trips and having spares
in case of breakdowns, accidents or operator absences.
Literature Review
These causes are generally environmental in nature. While the availability of vehicles
and drivers tends to be driven by operating policies and can be controlled to some
extent, random externalities still exist, such as vehicle accidents and driver sickness.
The study identified the following as significant inherent causes of unreliable service
considered to most significant:
-
Initial deviation from scheduled times or scheduled headways caused by traffic
conditions, late departures from origin points (garage or terminals), or uncommon
volumes of passenger boardings and alightings. Such deviations tend to propagate,
creating unbalanced loads and may worsen conditinos downstream.
-
Variation in speed (travel times) between consecutive buses caused by exogenous
factors and operator behavior
-
Unrealistic scheduled running times and recovery times that buses are unable to
follow.
-
Operator behavior and inadequate supervision that impedes proper headway control
or schedule adherence, especially at terminals.
2.3.5 STRATEGIES TO IMPROVE RELIABILITY
The study categorizes strategies to improve reliability as follows:
1)
Priority strategies, where transit vehicles receive special treatment to reduce the
influence of external factors.
2) Control strategies, which involve direct handling of active service operations.
3) Operational strategies, that relate to changes in route, schedule and resource
allocation.
The authors also group strategies into two other categories, according to their
application:
a) Preventive strategies, aimed at reducing the likelihood of reliability problems
developing.
b) Corrective or restorative strategies, directed at avoiding further propagation of
problems and restoring normal operations.
Strategies identified by the study are presented in Table 2-6.
33
Chapter 2
Table 2-6. Strategies to Improve Service Reliability
Preventive
Priority______
Corrective
___
Exclusive Lanes
X
Signal Priority
X
Control
Holding
___
X
X
Passing / Overtaking
Turnback
X
X
Skip stops
Speed Modifications
X
X
X
Reserve vehicle and drivers
Schedule adjustments
X
X
X
X
Express service
X
Operational
Improve vehicle access (fare collection,
boardings/alightings)
*
X
Source: Transit Reliability Study. (Abkowitz et al. 1978)
Previous studies evaluated the effectiveness of these strategies through empirical
analyses and simulation models. However, the authors argue that these evaluations
measured improvements in reliability by changes in mean values rather than other
measures of reliability.
The previous studies showed that priority strategies, such as exclusive lanes and signal
preemption, had the potential to reduce mean travel time. On control strategies, the
review of studies found a reduction in mean travel and wait times due to roadside,
central control or automated vehicle monitoring. Holding strategies were shown to have
a positive effect on headway regularity, but a negative impact on travel times.
Turnbacks and skip stops were shown to have negative effects on wait times and
transfers. A full cost-effectiveness evaluation would be required to determine whether the
extra resources needed for vehicle monitoring could be justified. For operational
strategies, there was an emphasis on good schedule planning to reflect predictable
variations and the use of the reserve fleet.
The authors conclude with a number of future research proposals to supplement and
extend our understanding of service reliability. Research should focus on the
relationships between reliability and behavior, for both travelers and operators, and the
evaluation of service measures to improve transit reliability. Recommended studies
include: 1) empirical models and simulation tools to determine the causes of unreliability;
2) evaluation of reliability measures and practical testing of those proposed in the study
to examine appropriateness and cost-effectiveness; 3) development of relationships
between reliability and travel behavior, and 4) the effects of unreliability on the cost of
operations. Future studies should also focus on the implementation and effects of
strategies to improve service reliability, such as control point holding, transit priority
schemes, route restructuring and schedule planning.
34
Literature Review
2.4
PORTLAND, OREGON STUDIES
A number of studies conducted by Portland State University, in conjunction with TriMet,
the transit provider for the Portland, Oregon metropolitan area, have highlighted the
innovative applications of TriMet's Bus Dispatching System (BDS) to improve bus
service. While other studies have looked at the use of AVL and APC systems to
improve service reliability, particular attention is given to the TriMet studies because of
their depth and quality.
One of the advantages of the BDS system is the integration of AVL and APC
technologies to collect detailed stop-level information. This has supported both real-time
and off-line applications, increasing the pool of potential users of the data and the
potential opportunities in monitoring service, measuring performance, and schedule
planning. The system is also widely deployed with AVL on all buses and APC installed
on 72% of the buses (Strathman et al. 2004), which eases the sample size constraint.
2.4.1 IMPROVEMENTS
IN RELIABILITY
One of these studies (Strathman et al. 1999) documents improvements in reliability and
running times after the implementation of their AVL and Computer-Aided Dispatching
(CAD) technologies. The study compared key indicators of service reliability, on-time
performance, headway adherence and running time variations, for eight bus routes of
different types (radial, cross-town and feeder). The service reliability measures were
standardized (normalized with respect to the mean) for direct comparison given the key
route characteristics (route type, service frequency, time of day).
The statistical results comparing the baseline and follow-up period showed an overall
increase in on-time performance, from 61.4% to 67.2% of all trips, with improvements
concentrated in the AM peak period (an increase of 129%). A closer look at the
distribution of delay revealed that it had shifted to the right, with a decrease in early
arrivals (by almost 37%) and an increase in late arrivals (by more than 14%). This
indicated that although more trips were arriving at their destination within the window of
one minute early and five minutes late, the average delay had actually increased.
The coefficient of headway variation throughout the day decreased by almost 5%, with
improvements concentrated mainly in the PM peak (a reduction of 15%). The share of
normal headway trips (those with headway ratios 3 between 70% and 130%) increased
by only about 1 %, while the cases of bus bunching (headway ratios below 70%)
3 Headway
ratio was defined as:
Headway Ratio (HR) = (Observed Headway / Scheduled Headway) * 100
Source: Strathman et al. (1999) "Service Reliability Impacts of Computer-Aided Dispatching and
Automatic Vehicle Location Technology: A Tri-Met Case Study"
35
Chapter 2
decreased by 15%. Overall estimated excess wait-time 4 also decreased (nearly 7%) in
the follow-up period, concentrated in the PM peak period.
Mean running time did not change significantly, but its coefficient of variation declined by
18%, concentrated mainly in the AM peak inbound trips. Comparing running times of the
baseline and follow-up period, an increase (+12%) was observed in the percentage of
buses completing their trip within +/- 7.5% of their scheduled run time.
Making use of the APC data, Strathman et al. also developed a bus running time model
to control for the effects of route design and passenger demand and allow for a more
careful examination of the effects of the BDS. The Ordinary Least Squares (OLS)
regression model estimates running time as a function of departure delays, number of
stops, route length, boardings and alightings, scheduled headway, peak periods and an
After-BDS dummy variable. The parameter estimates of the model are presented in
Table 2-7.
Table 2-7. Parameter Estimates for Running Time Model
Variables
Units of Variable
5.19
Constant
Departure Delay
Parameter Estimates
Observed minus scheduled departure time, in
- 0.30
minutes, at terminal
Stops
Number of APC-recorded passenger stops in
2.90
the trip
0.34
Distance
Route length, in miles
Boardings
Total passenger boardings in trip
0.01 *
Alightings
Scheduled Headway
Total passenger alightings in trip
Scheduled headway, in minutes
0.01 *
AMin
Dummy variable: 1 for inbound AM peak
- 1.41
- 0.13
period trip, 0 otherwise
PMout
Dummy variable: 1 for outbound PM peak
3.70
period trip, 0 otherwise
After BDS
*
Dummy variable: 1 for observations after BDS
- 1.45
implementation, 0 otherwise
at the 0.05 level.
insignificant
The t-ratios are
** Source: Service ReliabilityImpacts of Computer-Aided Dispatching and Automatic
Vehicle Location Technology: A Tri-Met Case Study. (Strathman et al. 1999)
4 Excess
wait time was defined as:
Ex. Wait Time (EW) = ((VarianceHR / (2 * MeanHR)) / 100) * Mean Observed Headway
Source: Strathman et al. (1999) "Service ReliabilityImpacts of Computer-Aided Dispatching and
Automatic Vehicle Location Technology: A Tri-Met Case Study"
36
Literature Review
The authors make the following comments on the results of the model:
-
Under the assumption that schedules allow for buses to make up some of the initial
delay along the route (by providing more than the average running time between
each timepoint - a somewhat risky strategy which may encourage early departures
downstream), the authors expected the "Departure Delay" parameter to range from
zero (none of the delay is made up) to minus 1 (all of the delay is made up). The
model affirmed this expectation and estimated that operators recover about a third of
their initial delay. (Note: this seems somewhat counterintuitive, as one would expect
running times to increase with late departures, due to the likelihood of an increased
number of passenger boardings).
-
Boarding and alighting coefficient estimates were not significant, implying loads are
within the vehicles' capacity limitations and do not significantly influence running
times.
-
The authors expected scheduled headway to be inversely related to running times on
the basis of a greater need for control actions, such as holding, on higher frequency
routes which increases travel time. This expectation proved to be correct as the
model estimates a decline in running times with increased headway. However, a
more general view would attribute this relationship to larger headways tending to
have lower running times because of lower demand, and not the need for control
actions.
-
The authors expected higher running times during peak periods (positive model
parameter estimates for the AMin and PMout variables). This expectation is
contradicted in the AM peak period, where running times were estimated to be
slightly lower (by almost 1.5 minutes). The PM peak period does show increased
running times (by nearly 3.7 minutes) compared to other trips, showing congestion
problems are worse later in the day.
-
Running times for AM peak inbound trips were estimated to be slightly lower (-1.5
minutes) compared to other trips in the day. The parameter estimate for inbound AM
peak period trips also contradicted the authors' expectations of higher running times
during peak periods due to traffic congestion.
-
After controlling for the effects of the other variables, the model estimated that the
BDS system reduces trip running times by almost 1.5 minutes on an average route.
Benefits from the initial impacts of the BDS were analyzed as having cost-saving
potential.
In general, average running times were shown to be slightly higher in the follow up
period. Interpretation of the results was that the reductions in running times due to the
BDS have been masked by the effects of changes in other operational characteristics,
such as an increase in average number of stops, in scheduled headways and in average
departure delays.
37
Chapter 2
Strathman et al. organize benefits into three categories: 1) reductions in passenger wait
times, 2) reductions in passenger in-vehicle travel times, and 3) improvements in
operator running times. These translate to increased passenger satisfaction and cost
savings, and when valued using average passengers' wage rate and operational costs,
they yield savings up to $5.4 million.
The study concludes that the impact of the automated bus dispatching system were
positive, with improvements in on-time performance and in headway and running time
variability. The system has the potential to yield greater benefits through further use in
operations control and monitoring, and in service planning.
2.4.2 SCHEDULE EFFICIENCY AND OPERATOR BEHAVIOR
Another study (Strathman et al. 20011) uses archived data to analyze the efficiency of
schedules and investigate the effects of operator variability on running times. The
analysis consisted of comparing actual and scheduled running times and recovery times.
The analysis was applied to one route (14 Hawthorne) characterized as a radial route
with frequent service and heavy passenger loads, and external traffic conditions
including moderate-to-heavy traffic, on-street parking on most of the route, numerous
signalized intersections and random delays from openings of a river bridge crossing.
The analysis assumes the definition of optimal running time and recovery times
described by Levinson (1991). Levinson argues that "running time for the route should
be set at a value slightly less than the median/mean" and "appropriate recovery time is
defined as the difference between the chosen [running time] benchmark and the running
time associated with the 95th percentile trip in the frequency distribution." This ensures
that, under normal operating conditions, 95 percent of the trips will be able to maintain
their schedule. These "optimal" values were compared with their scheduled times for
five different time periods (early AM, AM peak, mid-day, PM peak and evening).
The results showed that scheduled running times are lower than "optimal" for all
outbound trips and that scheduled recovery times were higher than "optimal" for all trips
except in the PM peak inbound trips. Even though these scheduled times follow
standard practice of setting low running times with generous recovery times to avoid
excess time and late departures, the total scheduled (running plus recovery) times were
found to be excessive (higher than "optimal").
The authors also present a more detailed analysis of each scheduled trip on all bus
routes in the system to account for different route types and service patterns. The
analysis considers three different recovery benchmarks: 1) Levinson's optimal recovery
time based on the 9 5th percentile value, 2) based on operators' contract requirements,
recovery time equal to 10% of the median running time, and 3) the "rule-of-thumb"
standard of 18% of the median running time. The analysis used running time data from
281,305 trip-level observations from 5,479 scheduled daily trips of 104 bus routes over
65 weekdays.
38
Literature Review
The results of this analysis found that the average scheduled running time conforms to
Levinson's use of the median running time value. However, for all three recovery time
benchmarks, there are excesses in scheduled recovery time and in total scheduled
times, which includes running and recovery times. The latter were found to be in excess
by 7.30, 7.92 and 3.82 minutes based on the benchmark values by Levinson, contract
requirements and "rule-of-thumb" standard, respectively. Based on these excess time
values and a marginal operating cost of $42.00 per platform hour and 255 days of
weekday service, the analysis presents estimates of potential annual cost savings of
$7.1, $7.7 and $5.1 million dollars, respectively. Using the standard "rule-of-thumb"
value of 18%, 81 of the 104 routes had excess scheduled times. Limitations also exist in
reducing recovery times, such as clock face schedules, labor contracts, transfer
optimization and operator scheduling.
Nonetheless, the analysis suggests that adjustments towards a more efficient schedule
could potentially yield cost-reduction by more efficiently scheduling vehicles while still
providing the same level of service to customers. Also, the study clearly demonstrates
the potential of using automatic data collection technologies in scheduling and service
planning.
Another component of the study (Strathman et al. 20011) is a model that explored the
influence of operator behavior on running times. The model estimated run times as a
function of route length, number of stops made, time period variability, route type,
passengers boarding and alighting, scheduled headway, seasonal variability, and an
operator-specific dummy variable. Fifteen bus routes, of different types, were selected
and the study comprised of 110,743 valid weekday trips (49.7% of all scheduled
weekday trips for the selected routes) in the summer and fall sign-up periods. The
parameters estimates of the model are presented in Table 2-8 and the coefficients
represent the change in running time associated with a unit change of the given variable.
39
Chapter 2
Table 2-8. Running Time Model
Coefficient
Intercept
Distance (per mile)
Lifts (does not include 8.10 seconds to stop)
Stops (involving a single boarding / alighting)
Early AM (compared to mid-day)
AM Peak (cmpared to mid-day)
PM Peak (compared to mid-day)
Night (compared to mid-day)
Feeder (compared to radial route)
Crosstown (compared to radial route)
Peak Express (compared to radial route)
Ons+Offs
(in seconds)
573.09
206.00
59.80
8.10
- 256.66
- 99.84
138.43
- 248.04
- 418.47
- 506.14
- 1088.90
3.36
Ons+Offs2
- 0.0016
Headway (per 1 minute reduction)
- 10.76
Summer
- 26.40
Source: Evaluation of Transit Operations: Data Applicationof Tri-Met's
Automated Bus Dispatching System. (Strathman et al. 20011)
*
With regard to the operator-specific variables, the results of the model indicate that 70%
of the recovery time allocated (based on the standard of using 18% of the mean running
time) is needed to account for variation in operator behavior, and 17% of running time
variation can be attributed to operator behavior factors.
To further explore these factors, the authors specify a model to estimate the individual
operator effects as a function of experience, average delays and work assignment type
(trippers, extraboards and reliefs). The R2 value of the model is 0.09 and the estimated
parameters, along with the expectations, are shown in Table 2-9. Model parameters
showed a decrease in nearly 0.57 seconds in running time per trip for each additional
month (nearly 7 minutes for each year) of operator work experience (i.e, older drivers
tend to drive faster). The departure delay parameter was not significant, and of the work
assignment variables, only the tripper parameter was significant, estimated to increase
running time by 40 seconds compared to duty types. Based on the results of these
analyses, recommendations to reduce variability included operator training and
increased field supervision, and that the cost effectiveness of these strategies be
assessed.
40
Literature Review
Table 2-9. Operator Model Parameter Estimates
Variable
Expectation
Intercept
Mean Value
(Std- Dev.)
__________
Experience (months of
service)
Complaints (nurmber per
year)
Departure Delay (average
delay over sampled trips)
Regular Service
Tripper (1 if assignment is a
tripper, 0 otherwise)
Extraboard (1 if operator
has extraboard assignment,
0 otherwise)
Relief (1 if operator has
relief assignment, 0
otherwise)
lnversety related (increased driving skill, familiarity and
operating conditions)
Unclear in relationship of complaints and adherence
inversety related (makes up deay by running faster)
94.3
(88.7)
3.9
(4.5)
101.2
(54.9)
Coefficient
ft-value)
25.15
I1.07)
- 0.57
1- 5.69)
- 1A3
i-
.92)
0.19
g-1.49
-
Require more tme, typically filled by parl-4ime operators
during peak
Unclear, but filled by full-tLme operators
Require more, filled by newer operators
0.20
(0.40)
0.21
(0.41)
40.01
(1.78)
- 5.54
(- D.26)
0.18
(0.39)
29.8
(1.34)
* Source: Evaluation of Transit Operations: Data Applicationof Tri-Met's Automated Bus
Dispatching System. (Strathman et al. 20011)
2.4.3 HEADWAY DEVIATIONS AND PASSENGER LOADS
Another study by Strathman et al. (2003) investigates the relationship between headway
deviations and passenger loads using archived AVL-APC data. Deviations in passenger
loads will tend to contribute to a deterioration in the headway distribution due to
abnormally high or low dwell times. Simultaneously, extreme headway deviations will
create disproportionate passenger loads. The study included ten of TriMet's 99 bus
routes, providing cross-town (2 of the 10 routes) and radial services (8 of the 10 routes),
and was limited to weekday service for a period of 6 months. The mean peak passenger
load is normally lower than bus seat capacity, which is 43 for a standard 40 foot bus and
39 for a low-floor bus. However, the coefficient of variation in passenger loads for these
routes range from 0.30 to 0.40. Routes with high mean loads and high coefficient of
variation experience greater incidence of overloads. Scheduled headways for these
routes ranged between 5-12 minutes5.
A two-stage least squares regression model was formulated to account for the
simultaneity between passenger loads and headway deviations. The first-stage
41
5 Range of headways determined from posted schedules on TriMet's website: www.trimet.org (July 2005)
Chapter 2
equation estimated headway delay6 at the peak load point as a function of headway
deviation at the route origin (terminal point), operator experience and distance from route
origin to peak load point. The second equation estimated passenger load at the peak
load point as a function of headway deviation at the peak load point, scheduled headway
and a low floor vehicle dummy variable. The two-stage regression model meant that
instead of using the observed values of headway deviation, the estimates of headway
deviations from the first equation were used in the second equation.
The model results indicated passenger overloads were indeed primarily caused by
headway deviations. It is estimated that an increase of 1 minute in headway deviation at
the peak load point increases load by 2.6 and 2.0 people for the AM peak and PM peak,
respectively. Further analysis of the first equation estimates revealed that a one-minute
increase in headway deviation at the terminal point generally resulted in an increase of
approximately 45 second in headway deviation at the peak load point. Reductions in
headway deviations, especially at the origin (terminal), would yield significant reductions
in passenger load variation. The authors recommended that strategies to control for
delays should be focused on improved field supervision or, in the case where running
times and recovery times are insufficient, on schedule adjustments. The authors also
argued that this type of analysis with archived data provides agencies with the data and
tools to target potential problems and to implement control measures more efficiently.
2.4.4 OPERATIONS MANAGEMENT
Data availability at TriMet has also allowed analysis of archived data to improve
schedule and operations management (Kimpel et al. 2004). The paper was a summary
presentation of a number of performance reports used by TriMet to monitor service
quality and operational efficiency. A summary of the key findings highlighted in the
report are:
-
The detailed design of the TriMet BDS has allowed for the generation of various
reports analyzing different performance measures for multiple purposes.
-
Agency staff are able to customize reports to monitor schedule efficiency, passenger
loads and capacity, on-time performance and operator behavior, which can be
broken down by route, direction and time of day.
-
The potential exists to analyze the relationship between operator behavior and
service reliability. The reports can track operator performance through measures
such as percent of late departures and average minutes early or late. These allow
peer comparisons among operators and help identify the effects of operator
variability in order to improve schedules and supervisory actions.
Headway delays are referred to as positive or negative deviations from scheduled headways. A positive
deviation means a greater-than-scheduled headway, while a negative deviation signifies a smaller-thanscheduled headway. To avoid confusion, this study will refer to headway delays as "headway deviations"
6
42
Literature Review
In summary, the service delivery reports generated by TriMet provide a detailed
understanding of route performance in order to identify potential problems, investigate
the effects of various service attributes and serve as feedback and diagnostic tools to
improve scheduling, planning, and operations.
2.4.5 OPERATIONS CONTROL
In the area of operations control, Strathman et al. (20012) present a review of previous
operations control practice and an experiment to evaluate the potential of the TriMet
BDS to manage headways. The review shows that previous studies had analyzed
various strategies such as holding, short-turning, signal priority, but had been limited by
availability of data due to the large costs of data collection and supervisory restrictions.
Field supervisors are located at specific points and do not have easy access to
information about the whole system. The study examined the potential of APC and AVL
technologies to estimate the effects of real-time control strategies and implement them
more effectively. The experiment involved comparing passenger load variations and
headway variations for a baseline period and a control period, where various control
strategies were undertaken by field supervisors. The statistical analysis performed did
not provide significant evidence of the effects of the applied control strategies; although
the desired outcome of more balanced passenger loads was attained. The authors
conclude that the use of real-time APC and AVL for control purposes has been limited,
and there is a need to develop a decision support system to more effectively aid
dispatchers and field supervisors in maintaining and restoring service reliability.
2.5
RESEARCH POTENTIAL
The framework developed by Abkowitz et al. (1978) served as a theoretical discussion of
transit reliability with the objective of achieving greater understanding of reliability. The
framework recommends appropriate measures, identifies various causes of unreliability
and reviews the potential strategies available to deal with unreliability. It recognizes that
the characteristics of the distribution of service attributes give a better picture of reliability
than just the mean value of the measures, and that the evaluation of the mean and
variability combined is key to the understanding service problems. However, the
structure and conclusions of the study were largely theoretical and had limitations in the
analysis of real operating data to support its framework and apply these concepts in
practice.
The studies conducted in Portland have taken advantage of this potential and explored
the application of ADC systems to improve service planning, performance monitoring
and operations management. The extensive disaggregate data provided by TriMet's
Bus Dispatching System (BDS) have allowed analysts to study the impacts of AVL and
APC systems on service operations and performance, and the relationships between
certain service attributes. However, these studies do not provide a comprehensive
examination of the causes of service unreliability and their complex relationships with
service attributes.
43
Chapter 2
This research builds upon the strengths and weakness of the studies reviewed and
develops a practical framework to understand service reliability using archived AVL and
APC data. With the introduction of ADC systems, the limitations of high cost and
extensive manual labor involved with collecting and processing large data sets no longer
hinder the potential for more detailed analyses and a deeper understanding of reliability
problems. The data available from these systems provide the opportunity to extend our
understanding of unreliability and how service problems can be analyzed.
The practical framework presented in Chapter 3 aims to be as comprehensive as the
framework presented by Abkowitz et al. 1978, but based on the technological advances
in data analysis and application of decision support tools using archived data, as
described by Furth et al. (2003) and developed by TriMet and Portland State University.
Updates to the transit reliability framework include:
44
-
A review and selection of appropriate service measures that adequately characterize
service and help identify reliability problems. The effectiveness of the measures
presented by Abkowitz et al. (1978) is evaluated further using more extensive and
more detailed datasets.
-
An in-depth framework to identify the causes of service unreliability and evaluate the
complex relationship between them using archived data. These will go beyond those
presented by Abkowitz et al. (1978) to reflect the recent data analysis capabilities
available through automated data collection systems.
-
The type of analyses developed by TriMet and Portland State University are brought
together to analyze service reliability in a broader and more complex approach.
Rather than looking at individual relationships between different service attributes,
which is what these studies have evaluated, the framework attempts to take a step
further and provides more detail on the interrelationships between the causes of
service unreliability and the complexities inherent in their impact on service
operations. Analysis tools are developed to help separate out the various
contributions of interrelated service design and operations factors.
3. A FRAMEWORK FOR ANALYZING TRANSIT
RELIABILITY
Previous research has laid a strong foundation for the analysis of service reliability, but
gaps in understanding unreliability and its causes still remain. The implementation of
automated data collection systems presents an opportunity to expand previous research
and extend our understanding about reliability. The objective of this chapter is to
develop a practical framework to assess transit reliability, with an emphasis on using
data from automated collection systems.
The chapter is organized in the following manner. Section 3.1 reviews the need for a
practical framework. Section 3.2 presents the key characteristics of transit service
reliability. Sections 3.3 and 3.4 present the causes of service reliability problems and
potential strategies to improve reliability, respectively, while Section 3.5 describes the
relationships between these causes and strategies. Section 3.6 presents the proposed
reliability analysis process to assess service reliability, describing the steps to
characterize reliability, identify the causes of unreliability and apply appropriate
strategies to improve service.
3.1
FROM THEORY To PRACTICE
Prior research has clearly shown that service reliability is a key objective for both
travelers and transit agencies. It is clear that to efficiently implement strategies to avoid
reliability problems or reduce their impacts, the fundamental causes of unreliability must
be understood. However, it is clear that poor reliability can be triggered by many
different and interrelated factors and so it is difficult to fully understand and address the
underlying causes.
Abkowitz et al. (1978) acknowledged the need for further research to increase the
understanding of reliability, especially the need to develop empirical models to evaluate
their proposed measures and possible causes of unreliability. While the TriMet studies,
reviewed in Chapter 2, have begun to take advantage of new technologies to increase
our understanding of service reliability, a gap between the theoretical groundwork and
practical applications remains.
The 1978 study (Abkowitz et al.) was based on knowledge of transit reliability at the
time. Twenty-five years later, this research re-visits this framework to assess which
features are still appropriate and applicable in light of our improved understanding based
on better data.
The proposed framework includes:
-
The overall picture of service reliability including the characteristics of unreliability,
the causes of problems and potential strategies to improve service reliability.
45
Chapter 3
3.2
-
A review of data and possible service measures to characterize service unreliability.
The service measures consider the perspective of both travelers and transit
agencies, and what the trade-offs are between them.
-
The distinction between systemic and random variations in archived data.
Determining whether an observed variation, such as a long gap in service, is a
random occurrence or a systemic problem. It is important in data analysis to
determine the patterns of occurrence that help identify the underlying causes of
unreliability.
-
The development of performance reports and analysis tools that enable the transit
agency to assess current service and identify the causes of unreliability problems.
KEY CHARACTERISTICS
OF SERVICE RELIABILITY
This section presents an overall picture of service reliability, highlighting the key
characteristics of service unreliability and their impacts. A key aspect of transit reliability
is the distinction between low-frequency and high-frequency routes. The attributes of
transit reliability differ based on the frequency of the route because of the difference in
passenger behavior and passenger perception of unreliable service. Passengers on
low-frequency routes tend to time their arrival at a stop to catch a specific scheduled trip
and minimize wait time. Service will be perceived unreliable if their specific trip is not
running as scheduled, and they end up missing it or having to wait a long time because it
is late. Passengers on high-frequency routes tend not to consult posted schedules
because service is frequent enough they should not experience long wait times.
Unreliability for these passengers will be in the form of very long wait times (especially if
two buses arrive together after a large gap in service), and overcrowding.
As a reference, the Transit Capacity and Quality of Service Manual (TCQSM) considers
the threshold of 10 minute headways at which point the passengers consult the posted
schedules (low-frequency routes) or are assumed to arrive at random (high-frequency
routes).
3.2.1 ON-TIME PERFORMANCE
On-time performance is the measure of whether buses are operating as scheduled.
From the passengers' perspective, on-time performance is critical on low-frequency
routes, particularly for inexperienced travelers who base their departure time decisions
on printed schedules in order to minimize the expected wait time (Abkowitz et al. 1978).
When buses do not operate as planned, travelers experience poor service quality due to
increased wait times and higher likelihood of late arrival at destination. If a bus departs
early, travelers who time their arrival at the stop in order to minimize wait time may end
up missing the intended bus and have to wait a full headway for the next bus. If a bus
departs late, travelers experience longer wait times. Transit agencies experience
46
A Framework for Analyzing Transit Reliability
inefficient operations, possible increases in operational costs (to compensate for poor
service), and loss of patronage and revenues as a result.
The TCQSM argues that measures of on-time performance should be based on the
interest of the passengers which will vary by location on the route as well as the
scheduled headways. For most passengers, an early departure is not "on-time"
because if they timed their arrival based on posted schedule, they will have to wait a
full headway for the next bus. However, at points where more passengers are likely
to alight than board, an early arrival does not have as great an impact. This study
defines "on-time" performance as the range between 0 and 5 minutes late, with
consideration of whether to measure departures or arrivals based on the likelihood of
boardings and alightings. Table 3-1 presents the TCQSM proposed levels of service
for on-time performance, with "on-time" performance defined as 0 to 5 minutes late
applied to either arrivals or departures.
Table 3-1. On-Time Performance Levels of Service
Level of Service (LOS)
*
On-Time Percentage
A
95.0-100%
B
90.0 - 94.9 %
C
85.0 - 89.9 %
D
80.0-84.4%
E
75.0 - 79.9 %
F
< 75.0 %
Source: Transit Capacity and Quality of Service Manual -
2 "ndEdition
Benn (1995) presents a survey of transit agencies in North America that shows most
agencies define an early bus as 1 minute early and a late bus as 5 or more minutes.
The Chicago Transit Authority (CTA) defines on-time schedule adherence as buses
within 1 minute early and 5 minutes late for headways over 10 minutes. New York City
Transit (NYCT) uses the same 1 minute early and 5 minutes late window for on-time
performance (NYCT website, 2005). The MBTA's service delivery policy designates bus
service as on-time for arrivals and departures within 0 to 5 minutes of published
schedules (MBTA Service Delivery Policy, 1996)
3.2.2 HEADWAY REGULARITY
Headway adherence is often used to determine service reliability for high-frequency bus
service since passengers often arrive randomly and headway irregularity can affect both
expected waiting times and the variability of passenger loads. The variability of
headways causes travelers to perceive service as unreliable, especially if two or more
buses arrive in a platoon after a long gap in service. There is clearly a problem when
buses drive along the route one right after the other, meaning somewhere there is a gap
in service and resources are not being used efficiently.
47
Chapter 3
The TCQSM considers headway adherence level of service (LOS) grades based on the
coefficient of variation of headways and the probability of bunching (see Table 3-2). The
coefficient of variation of headway is the standard deviation of headway divided by the
mean scheduled headway. The probability is defined as the probability that a vehicle's
headway will differ from the scheduled headway by more than 50%.
Table 3-2. Fixed-Route Headway Adhrence Level of Service
Level of Service
(LOS)
Coefficient of Variation
of Headway
Probability
A
B
C
D
E
F
0.00 - 0.21
0.22 - 0.30
0.31 - 0.39
0.40 - 0.52
0.53 - 0.74
0.75
s1 %
10 %
s 20 %
33 %
s 50 %
> 50 %
Comments
Service provided like clockwork
Vehicles slightly off headway
Vehicles often off headway
Irregular headways, with some bunching
Frequent bunching
Most vehicles bunched
Source: Transit Capacity and Quality of Service Manual - 2"d Edition
Benn (1995) reports that 28% of the transit agencies surveyed use headway adherence
as a criterion for performance evaluation, but that attention to this criteria is increasing
due to the availability of data from AVL technologies. The CTA defines regular headways
as headways below a threshold value of 150% of scheduled headway (readings above
are considered gaps in service) and above an absolute value of 1 minute (buses with
headways lower than 1 minute are considered bunched). NYCT defines an "acceptable
limit" for headways based on bus wait assessment, given as no more than 3 minutes
greater than scheduled headway during peak hours, and 5 minutes during off-peak
hours (NYCT website, 2005). This means, that headway deviations above 3 (or 5)
minutes are considered unreliable because customers have to wait longer than
scheduled. For MBTA bus service, the service delivery policy defines a good service
adherence as 85% of all trips in a time period are within 1.5 (or 150%) of scheduled
headway, for service with scheduled headways lower than 10 minutes (MBTA Service
Delivery Policy, 1996)
3.2.3 PASSENGER LOADS
Service frequency is set to ensure demand is evenly distributed and buses do not
exceed a given load factor 7, a measure based on the bus seat capacity. Benn (1995)
shows that seventy-two percent of transit agencies surveyed use a benchmark
maximum number of standees as criteria in schedule design standards. This benchmark
is a percent of the number of seats. When passenger loads are high, passengers
experience a lower level of comfort, and boarding and alighting times generally increase,
along with the dwell time at stops. Overcrowding eventually leads to a more severe
problem when passengers are not able to board the first bus that arrives because it is
7 Load
48
factor is defined as the occupancy of a vehicle divided by its seating capacity.
A Framework for Analyzing Transit Reliability
full and have to wait for the second vehicle. This obviously increases the wait time of
passengers, the probability of late arrival at destination and the user's frustration.
3.2.4 OVERVIEW OF INTERACTIONS
Before discussing the causes of unreliability and strategies to improve service, it is
helpful to provide a broad picture of the interactions affecting reliability of transit service.
At a high level, Figure 3-1 shows how service unreliability develops and how the system
may be restored to scheduled or normal service. Trigger events are defined as
occurrences, whether operational or external, which cause an initial deviation of a
vehicle from its scheduled time or headway. Many of these causes can also propagate
unreliability once deviations have occurred. The strategies to improve service reliability
are categorized as preventive and corrective, according to their application. Preventive
techniques aim to avoid the occurrence of problems and maintain regular service, while
corrective strategies aim to return service to normal and minimize the effects of
deviations.
Figure 3-1. Service ReliabilityInteractions
Preventive
(Type 1)
CorrectiveELIABI
(Type 2)
Normal
Cause:
De iations
Service
Trigger
from Schedule
The relationships between the different components are simplified in this diagram, since
the interrelationship of service attributes, causal factors and potential strategies are quite
complex. These complexities are discussed further in subsequent sections.
3.3
CAUSES OF UNRELIABILITY
This section describes the most significant causes of service reliability problems. The
goal is to provide further understanding of why they are considered significant causes,
how they impact service and how they are interrelated.
3.3.1 SCHEDULE DEVIATIONS AT TERMINALS
Schedule deviations (actual departure time minus scheduled departure time; thus, late
departures are positive deviations while early departures are negative deviations) can be
measured at any point along a route. Some buses may be able to make up a schedule
49
S----
---.---.--
--
-q
-
Chapter 3
deviation and return to their scheduled times further down the route, but others may not
be able to. A schedule deviation might propagate, with the vehicle falling further from
schedule as it proceeds along the route, leading to further service deterioration.
Emphasis is given to schedule deviations at the terminal because operators and
supervisors have more control at these points than at intermediate points along the
routes. Supervisors at terminals are able to monitor driver behavior and more
disciplined behavior is likely to result. Recovery times at terminals are intended to allow
buses to recover from earlier delays (i.e. late arrival on the previous trip) and still begin
the next trip on-time. Recurrent late departures from terminals may be an indication of
inadequate training, poor enforcement or supervision, or poor scheduling.
Small deviations from schedule at intermediate points are not as significant because the
scheduled times are based on average running times and variability is known to occur.
These deviations may also be the direct result of deviations at the beginning of trips from
which drivers are unable to recover.
Problems at terminals are also significant because unreliability tends to propagate
downstream, and any deviations from scheduled times or headways at the start of the
trip will affect a greater number of passengers.
Figure 3-2 illustrates the effects late departures from terminals could have on service.
Deviations have the potential to cause an increase in boardings at stops further
downstream, if passengers arrive randomly at downstream stops, as is often the case
with high frequency service. Increased boardings at stops result in higher dwell times,
which increase total running times. If not enough recovery time is available at the end of
trips, late arrivals of previous trips might carry over to the next trip and cause further late
departures.
Figure 3-2. Effects of Late Departure from Terminal
Late
departur
derure
1frerrmnal]
-
Increased
Increased
Increased
boardings
dwell times
running time
0 Late arrival
Not
enough
N recovery
ecoery
at terminal
An important concept here is the concept of random incidence when dealing with the
relationship between headways and passenger loads. If passengers are assumed to
arrive at random, which is the case for high-frequency route, the probability of a
passenger arriving in a long headway is greater than that of a shorter headway.
Therefore, if headways become imbalanced such that consecutive buses have
alternating headways of 5 and 15 minutes, the probability of a passenger arriving during
the 15 minute headways is 0.75. Thus, imbalanced headways will create an imbalance
in passenger loads between consecutive buses.
50
A Framework for Analyzing Transit Reliability
The relationship between headway deviation and passenger boardings is illustrated in
the following example. Assume a route with running time of 5 minutes between stops,
and vehicles departing from the terminal every 5 minutes. Passengers are assumed to
arrive at a rate of 2 passengers/minute at each stop, each passenger takes 5 seconds to
board the bus and there are no alightings at any of the stops. Table 3-3 presents the
resulting differences if one bus departs the terminal one minute late.
Table 3-3. Effects of a Late Departure
Scheduled Service
Bus_
Bus#1
Bus #2
Bus #3
Bus#4
ftm
Termil
3:00:00
3:05:00
3:10:00
315:00
A
Demt
Bus #1
Bus#2
Bus #3
Bus #4
on
Boardi
Bus
_
3:05:00
3:10:00
3:15:00
3:20:00
3:06:50
3:10:50
3:15:50
3:20:50
_BustA_
ftm
Temil
3:DD:00
3:06:00
3:10:00
3:15:00
___
BUS
pA
A
Dmld
10
10
10
10
.
10
10
10
10
Der
Boanas
_
3:10:50
3:15:50
3:20:50
3:25:50
3:11:40
3:16:40
3:21:40
3:26:40
10
10
10
10
10
12
7
10
on
a
A
Dwaft
10
12
7
10
Aime
Bus
__s
20
20
20
20
3:21:40
3:26:40
3:31:40
3:16:40
_
Boardn
o
A
D
3:10:50 3:11:40
3:17:00 1 3:16:05
3:20:35 3:21:05
3:25:50 3:26:45
10
13
6
11
20
25
13
21
Bdins
_on
11
9
11
9
31
29
31
29
3:17:35
3:22:25
3:27:35
3:32:25
Bus SC
Dmirt
B
BKs
3:05:00 3:05:50
3:11:00 3:12:00
3:15:00 1 3:15:35
3:20:00 3:20:50
Bus Stop C
B
Actual Service
Bus_
us
A
Boardnas
Arive
p
Boardings
Ie
3:16:40
3:23:05
3:26:05
3:31:45
3:17:35
3:24:10
3:26:25
3:32:50
11
13
4
13
31
38
17
34
Bus #2 leaves the terminal one minute late, and begins its trip with a preceding headway
of 6 minutes, not 5 minutes as scheduled. Because of the longer headway, there are
more passengers waiting to board this bus at Stop A, increasing the dwell time and
load on-board. In the mean time, Bus #3 leaves on-time with a 4-minute headway, and
has a lower dwell time because fewer passengers have arrived at each stop. By the
third stop, the headways have diverged from the scheduled 5:00 minutes to 6:35
minutes and 2:15 minutes for Buses #2 and #3, respectively. As the buses continue
along the route, Bus #2 headway will increase, leading to crowding and falling even
further behind schedule. Bus #3 will pick up fewer passengers, its headway will
decrease and eventually it will catch up with Bus #2. Of course, this example is highly
simplified. Passenger arrivals at stops are not a constant deterministic process,
boarding and alighting times are not linear and are affected by other factors (for
example, crowding and wheelchair accessibility), and other externalities, such as traffic
conditions, cause other variations in running times between stops.
Late departures from terminals can be the result of various factors:
-
Late pullout from the garage. A garage pullout is the start of a vehicle block, as
buses deadhead from the garage to the route terminal. Any deviations from
scheduled pullout times have the potential to carry over to the first revenue trip.
Late pullouts from a garage may be due to:
51
Chapter 3
A. Operator or vehicle unavailability, which includes vehicle malfunctions or
operator re-assignment due to absenteeism.
B. Operator behavior, which involves operators reporting late for work or delays
in starting workpieces (i.e. disregard for schedule).
-
Late arrival from previous trip. If recovery time at the terminal is insufficient, delays
from a trip will propagate to the next trip. Late arrival from the previous trip may be
the result of poor scheduling (inadequate schedule running time), major delays due
to traffic conditions or passenger boardings, or operator behavior.
-
Operator behavior. Late departures may occur if operators disregard scheduled
departure times, take a break time greater than the available recovery time, or report
late for duty when relief occurs at the terminal.
3.3.2 PASSENGER LOADS
Poor service quality is experienced by passengers when buses are highly loaded and
crowded. Many passengers find it uncomfortable to stand for a long time or they are
unable to use their travel time productively (Transit Capacity and Quality of Service
Manual). Transit disutility increases when passengers compare feeling cramped in a
bus with the privacy and comfort of a single seat ride on a personal vehicle. Service
quality also deteriorates with higher crowding levels because it becomes more difficult to
move around inside the bus. The added interference increases the dwell times at bus
stops as buses must wait longer for people make their way in and out.
Two passenger boardings conditions can affect service reliability:
-
Abnormal loadings. These tend to affect the load on one vehicle and can disrupt the
headway distribution. When one bus spends a significant amount of time at a stop,
the preceding bus has the chance to increase the gap between them as it proceeds
along the route, and the following bus is more likely to catch up with the first bus. A
greater number of passengers will have accumulated at stops downstream because
of the larger headway, so the bus will continue to fall further behind schedule.
Eventually, this leads to bunched vehicles separated by large gaps in service.
Abnormal deviations in passenger loads thus tend to initiate schedule deviations.
Take the example in the previous subsection. The same assumptions of passenger
arrivals and travel times hold, except this time all buses depart from the terminal ontime but a group of 10 passengers arrive at stop A at 3:07:00. Other passengers still
arrive at the stop at the assumed rate of 2 passengers per minute. The results of
this random passenger load, shown in Table 3-4, are increases in headway and load
variability.
52
A Framework for Analyzing Transit Reliability
Table 3-4. Effects of an Abnormally High Passenger Load
Scheduled Service
Bus Stp B
Bus St&p A
Depart
.fM
Bus#1
Bus#2
Bus #3
Bus #4
3:00:00
3:05:00
3:10:00
3:15:00-
Aag
LkSar
Rnardins
3:05:00
3:10:00
3:15:00
3:20:00
3:05:50
3:10:50
3:15:50
3:20:80
10
10
10
10
Bus
Denad
.ft=
T
Bus#1
Bus #2
Bus #3
Bus #4
main
3:00:D
3:08:DD
3:10:00
3:15:00
rha
Dar
t
ld
go
10
10
10
10
3:05:50
3:11:5
3:15:40
3:20:80
3:10:50
3:15:60
3:20:50
3:25:80
pad
d
mM
3:11:40
3:16:40
3:21:40
3:26:40
10
10
10
10
20
20
20
20
baa
3:16:40
3:21:40
3:28:40
3:31:40
Actual. Service
Bus Stop B
A
Boardn
Lad
w
_I
3:05:00
3:10:00
3:15:00
3:20:00
Auia
Bus StopC
Waild
&ha
Depad
_
__
10
tO
22
22
8
10
8
10
3:10:50
3:16:55
3:20:40
3:25:50
aawcing
3:11:40
3:17:45
3:21:25
3:26:45
10
12
7
11
Dpari
Boardirog
3:17:35
3:22:25
3:27:35
3:32:25
11
9
11
9
Bus
ad
m
a
Ai
20
34
15
21
3:18:40
3:2245
63:28:25
3:31:45
epWa
.Lod
D
31
29
31
29
t-pC
.Lad
M
Board=
I___
_
3:17:35
3:23:45
3:26:55
11
12
3:32:50
13
6
31
46
21
34
The scheduled headway of 5:00 minutes between buses diverges to headways of
6:10 minutes for Bus #2 and 3:10 minutes for Bus #3. Bus #2 will continue to pick up
more and more passengers down the route, falling further behind schedule and
increasing its headway. At the same time, Bus #3 will have less passengers arriving
and waiting at stops, have lower dwell times and will catch up with Bus #2, creating
the bunching effect of two vehicles platooning together and also creating large gaps
in service.
-
Inadequate frequency to meet demand. When actual demand is greater than
expected, buses will run with higher than planned load factors which can lead to:
A. Crowding. Overcrowding decreases the level of comfort, slows down
operations and may lead to passengers being unable to board the first bus
that arrives. When this situation occurs, travelers may feel frustrated in
having to wait another full headway for the next bus, especially after already
waiting a long time, or having to seek an alternate mode of transportation.
B. Poor on-time performance. The higher boarding and alighting volumes make
dwell times at bus stops greater than those scheduled. Vehicles fall behind
schedule and service is not provided as promised.
In both cases, it is clear that any deviations in passenger loads disrupt the headway
distribution, and problems tend to propagate. Even when passenger loads are light,
buses might tend to run off-schedule because of the decreased time spent at stops or a
decrease in the number of stops serviced. If buses are not required to serve all stops,
only those requested by passengers, there is a time difference "saved" between the time
it takes to drive by and the time it needs to decelerate, serve passengers and accelerate
back into service.
53
Chapter 3
3.3.3 RUNNING TIMES
The variability of running times, especially between consecutive buses, affects the
headway distribution. Transit planners acknowledge such variability in running times
and therefore schedule recovery time at the end of each trip to ensure subsequent trips
have a high probability of on-time departure. If running times are inadequate, the
majority of the buses will have a poor on-time performance.
When actual running times are lower than scheduled, buses will tend to run "early" for
most of the route. Passengers who time their arrival at stops according to schedules will
be left waiting for a bus that has already left, and will have to wait a lot longer than
expected. If drivers are instructed to hold at certain time points and depart only at set
scheduled times, passengers experience increased in-vehicle times. With very slack
schedules, transit agencies then fail to benefit from the possibility of higher frequencies
with the same level of resources, or the opportunity to reduce the total number of buses
needed.
If too little running time is scheduled, buses will not run as scheduled. The system would
have poor on-time performance, and passengers who time their arrival at stops to
minimize wait time end up experiencing longer wait times and lower probabilities of ontime arrival at their destination. Buses are also likely to arrive "late" at terminals, and if
not enough recovery time is allocated, late departures from terminals can result and
reliability problems will propagate throughout the day.
Overall, the result of inadequate running times is poor on-time performance and
inconvenience to passengers.
Running times are affected by traffic conditions and other externalities, and by operator
behavior, which are evaluated in the next sections. Some operators tend to drive faster
and more aggressively, while others tend to be slower. The speed differentials due to
traffic conditions, intersections and operator behavior will cause deviations in headways
throughout the route. Recovery times and operator behavior are also related. If
recovery times are not adequate, drivers might tend to delay the start of the next trip.
3.3.4 ENVIRONMENTAL FACTORS
Environmental factors relate to traffic and the external context of operations (Abkowitz et
al. 1978, Levinson 1991). These exogenous causes are generally random in nature,
and cause variability in scheduled bus service.
-
54
Weather. Lower visibility levels, cautious drivers or poor road conditions during
inclement weather can reduce the speed of vehicles on the road and increase the
probability of accidents occurring. During bad weather, boarding and alighting times
tend to be above average. For example, when it is raining, passengers waiting for a
bus with open umbrellas will not pack together as tightly and will take longer to board
as they enter the bus and close their umbrellas.
A Framework for Analyzing Transit Reliability
Interactions with other vehicles on the road. Buses operate on roads, and generally
in mixed-use traffic. The presence of other vehicles on the road reduces travel
speeds, and high volumes of traffic create congestion. Interference occurs in
numerous forms, all of which disrupt normal service: vehicle movements, such as
lane changes or turning movements, street parking, illegally parked vehicles or
unexpected breakdowns and accidents.
Speed differentials. Intersections force buses to decelerate and accelerate, reducing
travel speeds and increasing total travel times. Signalized intersections add even
more variability because of the random nature of arrival at the traffic light. Buses
may be fortunate enough to catch the green light phase of the signal and easily
traverse the intersection without having to stop. But at other times they may be
unlucky to catch the beginning of the red light phase and have to spend extra time at
the intersection.
3.3.5 OPERATOR BEHAVIOR
As mentioned in previous subsections, variability in operator behavior affects on-time
departures and vehicle speeds. More generally, operator behavior can have a great
impact on running time and headway variability, and can affect reliability in the following
respects:
-
Driver Availability. Operator absenteeism can cause late departures or result in
missed trips when not enough spare drivers are available to cover the absences. It
has the greatest effects on reliability when operators disregard the agency's policies
and procedures on advance notification of absence and supervisors are left to find a
replacement at the last minute.
-
Running Time Variability. Speed differentials are one of the main contributors to
running time variability. These are the result of traffic conditions and stop-and-go
operations, as mentioned previously, and of an operator's driving characteristics. It
is unrealistic to expect that all operators will drive at the same speed, and differences
in personal characteristics are inevitable. Some operators tend to drive faster and
more aggressively, while others might be slower and more cautious. These
differences in behavior may be the result of differences in driving experience or
familiarity with the route.
-
Poor Performance. Bad operators may report late to work, take long personal
breaks or drive aggressively, and disregard the schedules. The result of reporting
late to work is the late departure of that bus from the garage or terminal. If operator
relief is made at mid-route, a late arrival from the succeeding operator also impacts
passengers' total travel time as the bus must wait. If operators value longer personal
breaks than those given as recovery times at terminals at the end of each trip, they
are likely to disregard the scheduled departure time, take their desired break and
depart late on their next trip. Operators also disregard scheduled times and
55
Chapter 3
headways when they deliberately catch up with their preceding bus and ride along
bunched in order to pick up fewer passengers and have an easier trip.
Because operator behavior can have an impact on service operations, it is interesting to
analyze the effects of certain operator characteristics. Previous studies (Strathman et al.
2001, Kimpel et al 2004) identified the following operator characteristics as most
affecting their behavior and performance:
-
Years of experience. The longer operators have worked, the more experienced they
are with the vehicles and the routes. The general observation is that older operators
tend to drive faster than their younger counterparts because they are more
comfortable with maneuvering the bus, and are more familiar with the route. Years
of experience also factor in operator performance because senior operators may
choose better shifts and vehicle blocks, and have a better performance record
because they might tend to be happier at their jobs.
-
Operator Type. Differences in characteristics of full-time and part-time operators
may affect driver performance. Part-time operators tend to be less experienced than
full-time operators, and may be subjected to different labor union work rules that
could affect their overall performance.
-
Duty type. The type of duty may affect the performance of operators. Trippers are
short-duration runs, usually covered by operators working overtime. Performance is
affected by overtime work as it is not the regular scheduled work of operators and
they may be less familiar with the route operations or conditions. Relief and
extraboard operators may also be unfamiliar with the route, as it may not be their
usual route.
3.3.6 INTER-RELATIONSHIPS
BETWEEN CAUSES
It is important to recognize that the causes of unreliability tend not to be isolated and
independent. As stated earlier, an event triggers an initial deviation from schedule, and
further interactions tends to propagate the delay down the route. In many cases, the
effects of one cause yield the occurrence of another cause of unreliability. Figure 3-3
shows the various relationships between the identified causes of unreliability. These
interrelationships make it difficult to isolate the trigger event from the propagation effects.
56
A Framework for Analyzing Transit Reliability
Figure 3-3. Inter-relationshipsbetween Causes of Unreliability
"'om~in~dw.-I to---
Dp:tfom
IDrreguar
P30eengnr Loeds
Passe Loa
'
Inadequate
Running Time s
r
1]mial
Traffic Conditions
Ond ExlernItios
4~
________I
~'
O~eu~ri
Oprtr
-Uub pael for Sawt:lie,
pr
'5hodulm
it]
F;rl c 141
jksreI0
Puixt rom *
Q
W6
QO rag e
rDeviation
Unavailgbity
'~
Pulmlt
r~~t
Operalur
Behavior
As discussed earlier, deviations from scheduled times may be the result of poor
schedule planning or variability in service conditions. Small deviations in arrival and
departure times are expected because of variability in environmental factors, operator
characteristics and passenger demand. The impacts of deviations from schedule
depend on the frequency of service. For high-frequency routes, it is more the range of
variability in deviations than the absolute value of deviation that causes reliability
problems. If all buses are off schedule, but by approximately the same amount of time,
the headway between consecutive buses remains regular and there is no adverse effect
on passengers.
Operator behavior may be the cause that has the greatest effect on variability and
service unreliability. All drivers will operate buses slightly differently and thus create
variability in various aspects of service. Any negative deviations (late departures) will
cause an imbalance in passenger loads between consecutive buses, which then
propagate through to variations in running times. Externalities such as traffic and
weather will affect running times, as well as driver and vehicle availability in case of
accidents and difficulties reporting to work.
Taking into account these complexities in the causes of unreliability, a number of
recommended solutions are presented. The recommended strategies are targeted
towards preventing the causes of unreliability and correcting any deviations once some
of the underlying sources of problems are identified.
57
Chapter 3
3.4
STRATEGIES
This research will use the categorization presented in the 1978 transit reliability study
(Abkowitz et al. 1978) which considers two categories of strategies to improve service
reliability: 1) Preventive, aimed at reducing the likelihood of deviations from occurring,
and 2) Corrective or Restorative, directed at restoring service to normal once deviations
have occurred.
By identifying some of the possible causes of unreliability in bus service, transit agencies
are able to target their efforts to minimize the occurrence of problems. Preventive
strategies focus on reducing the variability of running times and dwell times, while
corrective strategies focus on reducing the negative impacts on passengers. The
strategies are evaluated based on the costs and feasibility of implementation, and the
trade-offs to operations and passengers.
3.4.1 PREVENTIVE STRATEGIES
Preventive strategies can be grouped into five different categories: route design and lane
priority, signal priority, operator-related, supervision, and scheduling.
Route Design and Lane Priority
Exclusive bus lanes allow buses to operate without interference from general traffic.
When traffic volumes are high, operators have less control over travel speeds and have
a harder time pulling back into traffic after servicing a stop. Exclusive bus lanes,
therefore, reduce the variability in running times due to traffic conditions and allow for
greater control of speeds in order to adjust for early arrivals or late departures.
However, introducing exclusive bus lanes is not always feasible, since it involves major
capital costs in redesigning and reconstruction, and its implementation is constrained by
current street dimensions, surrounding development and traffic volumes. In addition,
exclusive lanes, unless fully protected or grade separated, are still subject to
intersections, turning movements and illegally parked vehicles.
Previous running time models (Strathman et al 1999, Strathman et al 2001) have shown
an increase in running time with respect to the route distance. Naturally, running times
increase as the buses traverse a longer route, but the variability in running time is also
expected to increase with distance as there is more chance for triggering events to occur
and for initial deviations to propagate.
The models in these studies (Strathman et al. 1999, Strathman et al. 2001) also show an
increase in running time with an increase in the number of bus stops. This increase in
running time, along with an increase of its variability, is attributed to the stop-and-go
factors of servicing a stop mentioned previously. Shorter routes or fewer stops as
improvement strategies are limited by network design issues, available resources and
service standards, such as maximum walking distance to bus service or maximum
number of transfers.
58
A Framework for Analyzing Transit Reliability
Signal Priority
Traffic signal Priorityaims to reduce running time variability by reducing delays at
signalized intersections. General traffic conditions and the random arrival at an
intersection with respect to the signal phase of the traffic light create variability in the
amount of time buses spend at intersections. As previously mentioned, buses may be
fortunate enough to catch all the green phases at intersections and whisk through
intersections without having to stop, while unlucky buses may arrive at intersections
during the red-phase of the signal and be forced to stop and wait. This running time
variability among consecutive buses disturbs the headway distribution, eventually
creating bunches and large gaps in service. Signal priority allows for greater control of
traffic flow at intersections, reducing the variability in time buses must spend waiting at
signalized intersections. Signal priority may also be used to speed up a bus that has a
large preceding headway or that is behind schedule. In this case, passengers directly
benefit from improved headway regularity and decreased travel times (Abkowitz et al
1978, Turnquist 1981).
There are two types of traffic signal priority: 1) absolute priority,where a green phase is
given to all buses regardless of whether they are running early, on-time or late; and 2)
conditional priority, providing a green phase only to buses that are running behind
schedule. Furth and Muller (2000) examined the impacts of traffic signal priority on
average delay, measured as the time difference between the actual crossing time and
the time it would take a typical unimpeded vehicle to go through the intersection, for both
transit buses and private vehicle traffic in Eindhoven, The Netherlands. The results
showed conditional priority did not cause significant changes to private vehicle traffic.
For buses, average delays were greater than with absolute priority, but lower than
without priority. Another study (Kimpel et al. 2005) examined running times,
on-time performance and passenger excess wait time with data from before and after
the implementation of traffic signal priority. The analysis results were mixed with an
overall improvement in average running times, but mixed outcomes at the individual
route level and by direction and time-of-day. On-time performance decreased as the
tendency for early arrivals increased, and headway variability and excess wait time
increased.
Operator-Related
Reserve operators and vehicles are intended to avoid missed trips or avert large
schedule deviations. The absence of a scheduled operator or vehicle creates a gap in
service, which can significantly decrease service quality. Extraboard personnel cover for
operators who are on vacation, who call in sick or who are absent without leave. A
reserve fleet of vehicles is maintained to replace a missing bus resulting from vehicle
breakdowns, accidents or other emergencies. Standby buses can also be used to
service a route where peak demand warrants extra capacity. These on-demand buses
help avoid overloads and schedule deviations that can propagate unreliability throughout
the route. The number of reserve operators and vehicles is important, and the
disadvantage of this strategy is the extra cost of having reserve operators and vehicles.
59
Chapter 3
If the number of reserves is too low, supervisors can either assign overtime work or
adjust frequencies to cover for absent operators or broken-down vehicles, which
increases operating costs and decreases service quality. If the number of reserves is
too high, resources are underutilized as operators are not working while still getting paid
and vehicles in the garage are not being used. A more detailed analysis of the trade-off
between reserve policies and service reliability is described in Abkowitz et al. (1978).
Operator training serves to minimize the impacts of operator behavior on service
reliability. While it recognizes that there will always be differences between operators
that will cause variability in running times, it aims to make drivers aware of early signs of
service deterioration and use their own best judgment to improve service quality.
Operator training includes, but it is not limited to, review of operational policies and
procedures, route familiarity, driving classes, emergency-situation programs and
educational courses to emphasize the importance of good on-time performance and
balanced headways.
Operator incentives and penalties also serve as a strategy to induce better service.
Operator morale affects overall work performance and driving behavior. Thus, service
quality can be maintained by keeping morale high and by giving operators incentives to
perform well. On the other hand, stricter enforcement of policies regarding absenteeism
and poor performance helps induce better service. Imposing penalties and disciplinary
actions on operators who have a low performance grade serves to inhibit negative
operator behavior. This includes disciplining operators who tend to report late to work,
have a habit of early departures or tend to form platoons with the preceding bus in order
to have an easier duty.
Supervision
Service supervision serves as a strategy to both prevent and correct service reliability
problems. Supervisors are able to monitor operations, especially at terminals, to ensure
buses run on schedule and with reasonable headway spacing. Supervisors keep an eye
on operator behavior, preventing operators from taking personal breaks longer than the
allocated recovery time or avoiding early departures. These actions help reduce the
number of service reliability problems triggered by late departures from terminals.
Supervisors also exercise control to return service to normal after service reliability
problems have occurred. They can instruct operators to speed up or slow down in order
to follow schedules, adjust headways according to current demand, or employ any of the
corrective strategies described in the following section.
The introduction of automated vehicle location systems enables supervisors to better
monitor and control bus operations. Before, road supervisors only had information on
the buses as they drove by their point on the route and could only obtain vehicle location
information by polling operators through radio communications. This limited the ability to
accurately evaluate the impacts of control strategies, which could lead to actually
worsening the headway distribution. With AVL technologies, real-time information on
vehicle location allows supervisors to evaluate the route and make better control
60
A Framework for Analyzing Transit Reliability
decisions. This improves the effectiveness of control and lessens the risk of negative
impacts on passengers.
Schedule Adjustments
Adjusting schedules to reflect actual conditions also serves as a strategy to maintain
reliability. This strategy is directed to avoiding inadequate running times triggering
reliability problems, as described in Section 3.3.3. Passenger demand and traffic
conditions will gradually change over time, causing the running time distributions to
deviate from those used during schedule planning when vehicles and drivers were
assigned. If these deviations become large enough, schedules become unrealistic. This
disturbs the headway distribution, and causes early departures and/or unproductive time
(if the schedule is too slack), or late departures, bunching and crowding (if the schedule
is too tight). Periodic adjustments to schedules enable transit agencies to more
efficiently assign available resources and publish schedules that better reflect actual
service. Thus, the benefits of this strategy are more reliable service for passengers and
more efficient operations for transit agencies.
3.4.2 CORRECTIVE STRATEGIES
The most common corrective strategies are reviewed in this section: holding,
expressing, short-turning and deadheading.
Holding
Holding is the control strategy of delaying a bus at a time point for a set amount of time.
It aims to correct for a bus running early or prevent buses from forming bunches.
Holding can be schedule-based to ensure on-time performance, or headway-based
to maintain even headways between consecutive buses.
Schedule-based holding involves instructing buses that arrive at check-points early to
wait until their scheduled departure time. The assumption is that, by keeping to
schedules, the buses will run with more even headways. Implementation is subject to
reasonable schedules that buses are able to keep and adequate supervision to ensure
buses depart on-time (Turnquist 1981). If schedules are slack, a large percent of the
buses will spend too much time idle, as they wait for their scheduled departure time, and
passengers will get frustrated with the increase in travel time. If schedules are tight,
buses will be running late most of the time and the holding strategy will be ineffective in
achieving on-time performance or maintaining even headways.
Scheduled-based holding is more appropriate for low-frequency routes, where travelers
tend to arrive at stops based on posted schedules to reduce their expected wait times.
This type of holding is also appropriate for routes with important transfer connections.
By holding a bus until its scheduled departure time, passengers are less likely to miss
connections because the second bus departed earlier than scheduled.
61
Chapter 3
Headway-based holding focuses on delaying selected buses to balance the headways.
It involves holding a bus so that its preceding headway increases and its trailing
headway decreases. The bus in front continues service down the route and moves away
from the held bus, while the bus behind also continues regular service and moves closer
to the held bus, evening out the space (and times) between consecutive buses when
applied to buses with short headways. This prevents the occurrence of bus bunching
and helps balance passenger loads. As a result, headway-based holding is more
suitable for high frequency routes, where passenger arrivals are independent of the
schedule and bus bunching is more likely to occur.
Headway-based holding also decreases average passenger wait time at stops. For
high-frequency routes, passengers are assumed to arrive randomly at bus stops and
rarely time their arrival based on posted schedules. For these passengers, Wilson et. al
(1992) references that the expected passenger waiting time is given by,
w=
- 1+
2
ov2(h)]
where w is the average passenger wait time, h is the mean headway and cov(h) is the
coefficient of variation of headway, the standard deviation of headway divided by the
mean headway. The equation shows that for balanced headways, the coefficient of
variation is small and the expected wait time is approximately half the mean headway,
and the expected wait time increases as the headway variance increases. Therefore,
applying headway-based holding to balance out headways between consecutive buses
decreases the aggregate passenger wait time.
Turnquist (1982), as cited in Strathman et al (20012), analyzes two types of headwaybased holding: "Single Headway", where a bus is held until its preceding headway is
equal to its scheduled headway; and "Prefol", which consists of holding a bus until its
preceding and trailing headways are approximately the same. The first strategy only
requires headway information of the current bus, while the latter strategy involves
knowing the location of the following bus. Turnquist (1982) also points out that for
headway-based control strategies, wait time savings are maximized when all headways
are known in advance (Strathman et al 20012).
For both types of holding, passengers on board experience longer in-vehicle travel times
and transit agencies absorb the cost of longer running times (Strathman et al. 20012).
Previous studies point out that holding strategies may produce minimal improvements,
even negative impacts, and should be implemented early on the route, applied at an
optimal point where on-board loads are low and passenger demand is heavy at
immediate subsequent stops (Bly and Jackson, 1974, Koffman, 1978, Turnquist and
Blume, 1980, as cited by Strathman et al, 20012). The authors also point out that the
next generation of operations control research should exploit the potential of APC and
AVL systems to provide valuable information to decision-makers regarding the
implementation of holding strategies.
62
A Framework for Analyzing Transit Reliability
A study by Wilson et al. (1992) evaluates the potential uses of automatic vehicle
information (AVI) as decision-support tools for operations control strategies on the
MBTA's Green Line. The authors found that control decisions by supervisors were
mainly based on experience and communication with operators, and that occasionally
these strategies actually increased aggregate passenger wait times. The authors also
found that the results and effectiveness of control strategies varied widely because of
the lack of real-time information available to supervisors and decision-makers.
Real-time AVL and APC systems can provide decision-makers with useful information
about the buses in service, and enable them to better evaluate the impacts of
implementing control strategies. These systems can provide decision-makers with a
broader view of the route. Knowing the headways and the passenger loads on a bus
allows decision-makers to assess whether it is beneficial to hold a bus, what time point
to hold at, what length of time a bus should be held, and approximately how many
passengers on-board will be impacted. Off-line data from these systems may also prove
useful in helping decision-makers evaluate the effectiveness of holding strategies.
Archived data may be used to analyze the impacts of control strategy decisions, and to
develop decision-support tools that enable supervisors to estimate the conditions under
which holding and other strategies are most appropriate. Boarding and alighting profiles
generated from historical data would be helpful to approximate the number of
passengers, both on-board and down the route, who would potentially benefit from
holding a bus.
Expressing
Expressing involves sending a bus to a stop further downstream and skipping (not
servicing) some, or all, intermediate stops. The objective of this strategy may be either
to split bunched buses or to close a service gap further downstream, both in an attempt
to balance headways and improve service past the end of the express segment. Wilson
et al. (1992) state the ideal scenario for expressing a vehicle is when it has a long
preceding headway and a short trailing headway, and passenger demand past the end
of the express segment is high. This should minimize the negative impacts on
passengers at intermediate stops because the next bus will not be far behind, and
maximize the benefits of lower wait times on downstream passengers.
There are three types of expressing, categorized according to whether and how
intermediate stops are serviced, as described below.
-
Full expressing. The bus is expressed to a point downstream without serving any
intermediate stops.
-
Limited stops. The expressed bus services only a limited number of intermediate
stops.
-
Alighting-only. The expressed bus does not pick up any additional passengers and
only drops off at intermediate stops.
63
Chapter 3
Table 3-5 summarizes the pros and cons of each type of expressing.
Table 3-5. Types of Expressing
Type of Expressing
Benefits
Penalties
Full Expressing
Reduced travel times to on-board passengers
Inconvenience of transfer to on-
whose destination is past the end of the express
segment
destinations are intermediate stops
Reduced wait times to passengers waiting at
stops downstream
Limited Stops
of the express segment
board passengers whose
Increased wait times for passengers
at intermediate stops
Reduced travel times to passengers whose origin
Inconvenience of transfer to on-
AND destination are an intermediate stop or past
board passengers whose
the end of the express segment.
destinations are skipped stops
Increased wait times to passengers
at skipped stops
Alighting Only
Reduced travel times to on-board passengers
Reduced wait times to passengers waiting at
I stops downstream of the express segment
Increased wait times to passengers
at intermediate stops
The trade-offs of expressing involve the passengers on board, passengers waiting
beyond the express segment, and passengers at intermediate stops. Decision-makers
must weigh the benefits to downstream passengers and to the overall performance of
the route, considering the negative impacts on passengers at intermediate stops and
those on-board affected by the extra transfer.
An expressing strategy involves selecting the intermediate stops that are skipped or
served with limited service, considering the ability of a bus to express over a segment
and pass regular-service buses, and informing passengers of the change in service.
Again, real-time information enables decision-makers to better evaluate the trade-offs
regarding the decision to express a bus. Expressing a bus requires knowing the location
of each bus on the route in order to assess the overall system. The location and load
data from automated systems provide more detailed information to help supervisors
decide whether expressing is appropriate, which of the three expressing strategies to
apply, which bus is the most fitting to express, how long should the express segment be.
Archived data may also provide support to supervisors by analyzing previous expressing
decisions and assessing passenger demand estimates through load profiles.
Short-turning and Deadheading
Short-turning involves directing a bus to end its current trip before it reaches the
terminus, and service the route in the other direction. This strategy is employed to return
a late bus to schedule, or when extra service is needed in the opposite direction,
whether it is due to higher passenger demand or large gaps in service. The ideal
scenario for short-turning a bus as a control strategy is, as described by Wilson et al
(1992), when the bus carries a small load, its following headway is low, and there is a
large gap in service in the reverse direction. This will minimize the negative impacts in
64
A Framework for Analyzing Transit Reliability
the current direction because the number of passengers forced to transfer is low and
those waiting at subsequent stops will still have regular service, and it maximizes the
benefits of better service to passengers in the opposite direction.
Short-turning is constrained by the ability of buses to short turn and serve the reverse
direction. Route design and street layout may limit the number of control points at which
buses are able to easily switch directions.
Deadheading involves pulling a bus from service and running it empty for a segment of
the route. Deadheading is one of the most common control strategy employed in US
transit systems, and if applied appropriately, it can be effective in reducing service
irregularity (Eberlein et al. 1999). It is similar to expressing except that the vehicle runs
without any passengers on-board.
Both strategies are similar to expressing as it instructs a bus to discontinue regular
service in order to provide capacity at another location. All three strategies
inconvenience passengers at stops that would have been served if the control strategy
had not been implemented. On-board passengers are also troubled by the need to
transfer and incur additional travel time.
3.5
RELATIONSHIPS BETWEEN CAUSES AND STRATEGIES
This section is focused on identifying the links between the causes of service unreliability
and the best strategy according to the source of the problem. Taking into account the
complex interrelations between the causes of unreliability described in Section 3.3.6, a
number of recommended solutions are presented for each cause. While corrective
strategies can be applied once the causes have triggered a deviation in service,
emphasis is given to preventive strategies. These are targeted at reducing the
occurrence of unreliability once the underlying sources have been identified.
Figure 3-4 shows that operator behavior related deviations are the result of
absenteeism, disregard for schedules and speed differentials. Potential strategies for
absenteeism are changes in policies regarding absenteeism, disciplinary actions and
extraboard size. Stricter enforcement and disciplinary actions, along with control
decisions by supervisors, are strategies for operators who disregard schedules. Speed
differentials may be addressed through operator training, better supervision and control
strategies.
Supervision serves as both a preventive and corrective strategy to ensure service is
running at an adequate level, and at the sign of any major problems, supervisors are the
decision-makers regarding the application of control strategies. Operator training is a
preventive measure to address operating policies and increase familiarity with the
vehicle and route. This is aimed at decreasing the driving speed variability among
operators and increasing focus on on-time performance.
65
Chapter 3
Figure 3-4. Operator Behavior Strategies
.22Urc
Potential §21ulon
Disciplinary actions
-j
Absenteeism_-+
Changes
in absenteeism
policies
Change in extraboard size
Better supervision, stricter enforcement of policies
Disregard for
Operator
Behavior
Disciplinary actions
Schedules (early
or late
departures)
Control Strategies to return bus to schedule:
Holding, Expressing, Short-turning, Speed-up I
Slow-down
Operator training
Speed
J.
Differentials
[
Better supervision
Control Strategies: Holding, Speed-up / Slowdown
Variability in running times is inevitable, and such, deviations from average running
times are accounted for in the schedule through the use of time points and recovery
times. This variability can be caused by randomness in traffic conditions, passenger
loads and operators. Figure 3-5 show the various strategies applicable to running time
problems. Problems with traffic conditions may be resolved with lane priority, signal
priority or better scheduling. Exclusive lane or signal priority will decrease the variability
and randomness in externalities. Periodic adjustments of the timetable are important to
define on-time performance, account for changes in externalities, and assess if the
provided service meets current passenger demand and is operationally efficient.
Passenger loads that affect running time variability can be addressed through schedule
adjustments to ensure service meets demand, and corrective strategies to deal with
passenger demand variability. Operator training and corrective strategies helps reduce
the variability of operator characteristics.
66
A Framework for Analyzing Transit Reliability
Figure 3-5. Running Time Strategies
Source
Potential Solution
Exclusive bus lanes
CITraffic
a
Conditions
Running Times
Passenger
H
Loads
Operator
BehaviorK
L
Signal priority
I
Schedule planning: adjusting running times
*
Corrective strategies
I
Operator training
"
Schedule planning: adjusting headways
Corrective strategies: speed up/slow down
As described in Section 3.3.1, deviations from terminals tend to propagate down
the route and disrupt the headways between consecutive buses. Deviations can be
attributed to a number of possible sources: late pullout from garage, late arrival from
previous trip, operator behavior and poor supervision. Figure 3-6 provides the potential
solutions to problem of deviations from terminal. Strategies aim at: 1) correcting the
schedule to account for the actual distribution of running times and required recover
times to ensure an on-time departure for the following trip, and 2) decreasing the
likelihood of operators missing or reporting late for a work duty. Corrective strategies
also include a revision of the policies regarding the number of available operators
(extraboard and overtime assignment) and vehicles (spare ratios and maintenance
schedules), and how the agency handles the absence of an operator.
67
Chapter 3
Figure 3-6. Deviation from Terminal Strategies
Potential Solution
gourgo
Late Arrival from
Previous trip
Li.
Schedule planning: additional running time,
recovery time]
Revision of reassignment process (operator
availability: how to deal with operators reporting
late for duty, number of extra vehicles and
Late Pullout from
operators)
Garage
Maintenance programs (vehicle availability)
Better supervision, stricter enforcement of policies
Late Departure
from Terminal
Better supervision, stricter enforcement of policies
Operator
Behavior
Poor Supervision
Li
Disciplinary actions (if reporting late for work duty)
-
[j.
Increase recovery time to account for personal
time
Stricter enforcement of "on-time" schedules
Supervisor training
Disciplinary actions for poor performance
Figure 3-7 illustrates the potential strategies related to the effects of passenger loads on
service reliability. Due to the level of randomness associated with passenger loads, the
main strategies available are: 1) correcting for any deviations in schedules due to
variability in loads and 2) adjusting the current scheduled times to account for expected
average passenger loads.
68
A Framework for Analyzing Transit Reliability
Figure 3-7. Passenger Load Strategies
Potential Solution
Corrective Strategies: Expressing, Short-tuming,
Extra Vehicle(s)
Random influx of
passengers_
Schedule planning: adjusting running times,
decreasing frequency
Low Demand of
Passengers
Passenger
Loads
Corrective Strategies: Holding
Schedule planning: adjusting running times,
increasing frequency
High Demand of
Passengers
-
Corrective Strategies: Expressing, Short-tuming,
Extra vehicles
While externalities are beyond the operational scope of bus service, there are a number
of strategies, mainly preventive, to offset the service problems that arise shown in Figure
3-8.
Figure 3-8. Externafities Strategies
Potential Solution
Traffic Volumes
H+
Exclusive bus lanes
Signal priority to increase flow rate
Maintenance program (to avoid frequent
breakdown)
Improving response time to accident to reduce
impacts on route
Accidents,
Breakdowns
Control strategies: expressing, short-turning
Extra vehicle or operator availability to replace or
offset out-of-service bus
Externalities
Rescheduling existing buses to balance headways
-j
Speed
Differentials
Reduce number of stops
(intersections
and stops)
Signal priority at signalized intersections
Weather
[I
Programs to improve:
- Emergency service during inclement weather
- Bus shelter protection
69
Chapter 3
3.6
PROPOSED RELIABILITY ANALYSIS PROCESS
While the previous sections described the most common causes and strategies related
to service reliability, this section describes the analysis process to use archived data
from automated data collection (ADC) systems to evaluate service performance and
identify and address reliability problems. The proposed process includes three blocks:
1.
Characterization of service reliability. This block includes describing key data inputs
and appropriate service measures, as well as outlining a number of performance
reports to summarize service delivery and evaluate service reliability.
2. Identification of causes. Considering the complexities and relationships between
causes of unreliability, the analysis looks at deviations from the terminal as a focal
point, inferring the causes of such deviations and evaluating their effects down the
route.
3. Selection of strategies. The last block involves identifying appropriate strategies
which target the critical causes of unreliability so as to improve service reliability.
3.6.1 CHARACTERIZING
SERVICE RELIABILITY
The first block of the analysis process describes five key elements that an agency
should carefully analyze in characterizing service reliability on a specific route:
a) Data inputs
b) Output calculations
c) Service measures
d) Threshold values
e) Performance reports.
The goals of this block are to give transit providers a guide to determine delivered
service quality from the reliability perspective.
A) Data inputs
Different types of data are available through automated data collection systems, as well
as other sources, useful in evaluating service reliability. This research identifies key
data needed to characterize service reliability. Data inputs can be organized into four
different groups related to the trips, timepoints, passengers and vehicle, as summarized
in Table 3-6.
Not all these data items will be available through all automated data collection systems
because the level of detail recorded and stored varies by system. A number of data
70
----------- ==
-
-
____
__
--
- ---
A Framework for Analyzing Transit Reliability
'items also require integration with other databases, such as schedules, operator work
assignments and payroll, for more detailed follow up analyses.
As described by Furth et al. (2003), the greater the level of spatial and temporal detail,
the more accurate will be calculation of running time and headway. If data is recorded
based on time intervals or exceptions, it may prove difficult to estimate the time at a
specific location in order to determine the headway. Running time and headway are
more difficult to estimate from location-at-time data and the resulting errors will be larger.
The data items described below are based on those systems that provide time-atlocation data. The name and format of each data item vary across systems and transit
providers, so typical variable names are adopted below.
Table 3-6. Data Inputs for Analysis Process
Data Group
Description
Data items
1. Trip
Identifiers for specific vehicle trip
- Date (calendar)
- Route
ID
- Trip ID
- Run ID
- Block ID
- Operator ID
2. Time Point
3. Passenger
Activity
4. Vehicle
Attributes related to time-at-location
- Time Point ID
data records
- Direction
- Scheduled time
- Actual Arrival Time
- Actual Departure Time
Information on passenger
boardings and alightings
Events relating to vehicle
operations
- Passenger Ons
- Passenger Offs
- Doors open/close (time stamp, time spent)
- Speed (maximum speed, zero speed, etc.)
- Mechanical failures or vehicle alarms
1. Trip Data
Each data record has a set of identifiers needed for subsequent analysis:
Date and Route ID are the most basic identifiers, identifying the day and route of the
data record.
Trip ID, Run ID and Block ID are data fields that identify the corresponding bus and work
assignment. The trip, run and block IDs are based on the scheduled duties, which are
typically integrated into the system from the schedule database. For bus service, a trip
is the journey of a bus on a route from one terminal to the other terminal. A round trip
consists of two individual trips. A run is comprised of various trips assigned to an
operator, and a block is comprised of various runs and trips assigned to one vehicle. A
block typically starts and ends at the garage.
OperatorID is also typically included in the ADC system record to associate a work
assignment with the specific operator doing the work that day (one to one relationship for
each unique Run ID). This data item is generally available on the on-board system
71
Chapter 3
through operator log-in features. Operators may also be able to sign-in daily vehicleblock assignments and automatically pull-up scheduled trip information. This makes it
easier to match the actual service data records with the scheduled assignments.
Rescheduling adjustments must be noted and entered correctly in the system.
More information on operator assignments can be obtained from garage records and
supervisor log sheets, as well as the payroll (human resource) database. Data such as
years of experience and other operator characteristics are often vital in transit service
analyses to examine the role that operator behavior plays in transit reliability.
2. Time Point Data
For time-at-location data records, the locations are defined by a set of time point
identifiers. Transit providers use time points to post scheduled arrival times and
estimated running times for route segments.
Time Point ID is the identifying number of each time point on the route. The time point
identifier may be unique for each time point of each route and for each direction, or time
points may be shared among different bus routes. In the latter case, individual data
records are identified by the combination of the time point, route and direction identifiers.
Scheduled time at time point is the time at which the given vehicle is scheduled to be at
a particular time-point. The information is based on the current timetable, which is either
matched with actual data during post-processing or included automatically in the onboard system before hand. The latter is preferred since it reduces error and it also
allows the system to serve as an Automatic Vehicle Announcement Systems (AVAS).
Typically, AVL systems include the current route and schedule information to enable
real-time tracking features such as indicating when a bus is running early or late. It is
often difficult to ensure the accuracy of operator assignments, especially if there has
been a reassignment, or other rescheduling. In some cases, it may also be hard to
ensure that the scheduled trip data is correctly matched with the actual trip. Scheduled
time point data is important for any reliability and other analysis of running times and
schedule deviations.
Actual arrival at time point is the time the vehicle arrived at the time point. There are
different ways in which this time-stamp is determined. Most AVL-systems record the
arrival time of a bus when it enters an area surrounding the Global Positioning System
(GPS) coordinates defining the time point. Care is needed to ensure the accuracy of the
time-point coordinates, and calibration is required to ensure the actual arrival (and
departure times are being recorded at the correct locations.
Actual departure time at time point is the time the vehicle departed the time point. As
with the arrival time, the actual departure time is usually recorded as the vehicle leaves
the area around the time point. These two data items, actual time point arrival and
departure time, are available with most AVL systems that record time-at-location data,
however, some systems may be less detailed and only record one timestamp at the
72
A Framework for Analyzing Transit Reliability
specific location. Having both an actual and departure timestamp allows for dwell-time
analyses if data on door open and door close is not available.
It is noted that for both actual arrival and departure times, the AVL system records a time
stamp as the bus enters and leaves an area around the time point and not an event
record of door open/close. If the bus does not service the stop, time records are still
created as the bus passes through the area (care is given that the area around the GPS
coordinates must be big enough to record the bus as it passes through if it does not pull
over to serve the stop).
3. Passenger Activity
Passenger Ons and Passenger Offs are the number of passengers that boarded and
alighted at a given time point (or bus stop) on the route. Passenger-activity records are
typically generated by APC systems and vary in detail according to the method of
capture. If an APC system is not available, farebox transactions can provide data on
passenger boardings (data on entries only, not exits). Typically, less detailed systems
only record summary counts of passenger entry and exit with a time stamp, without a
time point or bus stop identifier to fix the location at which the boardings and alightings
occurred.
Matching passenger-activity records with location can provide detailed information on
dwell processes as well as load profiles. Unless the APC system is equipped with
location data or integrated with an AVL (or AVAS) system, post-processing is needed to
match passenger count records and location data records, typically through the time
stamp value. Problems can arise if the APC system captures data at the stop-level and
the AVL system records data only at the time-point level. Thus, passenger activity
analyses can benefit from integration with other systems such as AVL, AVAS, and AFC.
The better integrated the systems are, the better reporting capabilities and analyses can
be developed.
Passenger counts also need to be balanced and calibrated to ensure accuracy. APC
systems can capture passenger boardings and alightings in a variety of ways: beam
sensors around the frame of doors, pressure mats on the floor, etc. Inevitably, there will
be errors in count records, depending on the system's level of sensitivity and ability to
accurately distinguish on and off movements, especially in crowded vehicles. Thus, it
becomes important to do post-processing to balance loads, ensuring that loads equal
zero when it is known the bus was empty (for example, at the terminal) and below the
capacity (it is impossible for a bus to carry more than a given number of passenger, both
sitting and standing). Calibration may involve manually collected data from point-checks
and ride-checks.
4. Vehicle Activity
Many AVL and APC systems create data records for certain vehicle activities. These
data records include times such as the time a vehicle is not in motion (stopped), the
doors open or close, and the speed of travel. More detail on these and related data
73
Chapter 3
items are provided in Furth et al. (2003, page 20-21). These data items have the
potential to give a better description of actual operations, and help identify the causes of
deviations. For example, the use of a wheelchair lift at a stop would explain a high dwell
time, the frequency of zero-speed (vehicle not in motion) records along the route could
give insight into traffic conditions, or mechanical alarms or flags would make it easier to
identify deviations from accidents and breakdowns.
B) Output Calculations
From time-at-location data records containing valid data, the following service attributes
are routinely calculated:
Actual running time for a trip is the difference in observed (arrival or departure) time
between two time points. It is calculated for each trip between any two time points on
the route.
Schedule deviation is the difference between the scheduled and the observed arrival (or
departure) time of a bus at a time point. It is calculated for each trip at each time point
on the route.
Dwell time is the time the bus spent at a time point. If time-at-location data is available,
it is the difference between the actual arrival time and actual departure time at a time
point. If events such as door open and door close are recorded, the dwell time can be
inferred to be the time the doors remained opened at a stop.
Headways are calculated as the time between successive bus arrivals (or departures) at
a time point. It is calculated for each trip at each time point, based on the arrival (or
departure) of the previous (or next) bus at the same time point. Thus, a bus has a
preceding and a trailing headway. Typically, the preceding headway is used for headway
analyses since it is this headway that affects passengers served by a particular bus trip.
The time used can be either the actual arrival time or the actual departure time,
depending on what the transit provider wants to evaluate. To transit providers, the
headway is simply the time difference between consecutive buses, and thus, it does not
matter much whether the arrival or departure time is used, as long it is consistent
for all vehicles. Passengers tend to be concerned with the time between the departure
of the last bus and the arrival of the next bus, which is not consistent with either the
arrival or departure headway. The differences are most pronounced when buses
experience long dwell times as a result of unusual passenger activity or holding actions.
Headway deviation is the time difference between the observed headway and the
scheduled headway.
Headway Ratio is the ratio of the observed headway to the scheduled headway, for an
individual trip. It compares actual service with promised service and is most useful for
high-frequency routes. Headway ratio is a measure of headway adherence, where
74
A Framework for Analyzing Transit Reliability
values greater than 1 indicate a headway greater than scheduled and values less than 1
indicate a headway lower than scheduled.
This is also a good measure to identify large gaps in service and bunches, with the
threshold value dependent on the transit provider's policies for high-frequency routes.
Thus, a transit provider may consider a gap any headway greater than 1.5 the scheduled
headway and a bunch any headway lower than 0.3 the scheduled headway.
Passenger Load is the number of passengers onboard the bus. Automatic passenger
counters typically make a record of passenger boardings and alightings every time the
doors open and close, which may not necessarily coincide with the time points in the
route (typically not all stops are time points). Thus, passenger loads, calculated as the
running total of passenger ons minus passenger offs, may be given for each record or at
each time point. There are also two passenger load values per stop (or time point): the
arrival load (passengers onboard when the bus arrives) and the departure load
(passengers onboard at departure).
The mean and observed coefficient of variation of passenger loads may provide a good
profile of the current demand distribution. These are also useful for service planning and
operations control decisions.
C) Service Measures
Service measures are the set of aggregate metrics used to characterize overall bus
service, measure performance and evaluate service delivery. Service measures are
needed to compare promised and actual level of service and are fundamental in
characterizing service reliability. They consist of summaries of the individual trip outputs
defined in the preceding section. Transit providers use them to assess current service
and determine both the current level of reliability and whether it is improving or
deteriorating over time.
Service measures are also useful both to identify causes and to select effective
improvement strategies. As described by Abkowitz et al. (1978), measures are
important to: 1) identify and understand reliability problems, 2) identify and measure
improvements, 3) relate improvements to strategies, 4) modify strategies, methods and
design to achieve greater improvements in reliability.
Abkowitz et al. describe the need for measures to accurately describe the variability in
service, and reflect its impacts on both travelers and transit providers. The Transit
Capacity and Quality of Service Manual base their recommended service measures on
the following criteria: 1) best represents the passengers' perspective, 2) easily
quantified, and 3) in current use by transit agencies.
Service measures for a route are calculated for a number of trips, grouped by time
period and/or time point.
75
Chapter 3
1. Distribution of Service Attribute
The distribution of a specific service attribute reflects the variability in that attribute and is
at the heart of assessing reliability no matter what the attribute of interest. While the
distribution itself has a great deal of information, it is often represented by statistical
measures in the interests of quantification for ease of comparison. Service attributes of
interest include running times, deviations from scheduled times, and deviations from
scheduled headways.
The running time distribution is a good measure of reliability and is key in determining
whether scheduled running times are adequate. The distribution should generally
be centered on the scheduled running time because timetables are often built around the
average running times. It should be analyzed for time periods having comparable
running times reflecting similar traffic conditions and passenger demand. The typical
time periods used are early morning, AM peak, mid-day, PM Peak and evening.
However, more detailed analysis, such as hourly running time distributions, are possible
with the large datasets available through automated data collection systems, leading to
finer definitions of time periods. .
The distribution of deviations from scheduled time is a good measure to evaluate ontime performance and analyze arrival (or departure) time variability. It is a measure of
how well service adheres to schedule and is most directly applicable to low-frequency
routes, where the focus on reliability in meeting the posted schedules. If individual trips
are typically early (or late), frequent passengers will adjust their arrival time to minimize
their expected wait time. However, if there is significant variation in the bus arrival time,
passengers are left frustrated because they find service unreliable, missing their
scheduled bus or having to wait a long time.
The distribution will depend on whether scheduled departures from time points are
strictly enforced. Some agencies, with the intention of maintaining schedules, do not
allow any early departures from time points with operators instructed to hold until
their scheduled departure time if they are running early. For this case, the expected
average deviation should be small, but the distribution will be skewed to the right
(towards late departures). On the other hand, for transit agencies that do not have strict
schedule discipline at time points, a normal distribution centered on zero is expected. If
the distribution shows a non-zero peak/average, unreliability may be the result of poor
schedules, which might need to be revised to reflect actual conditions.
The distribution of deviation from scheduled headways is a good measure of headway
adherence and service delivery on high-frequency routes, where the focus is typically on
even vehicle spacing to minimize expected wait time. This distribution reflects how well
buses are adhering to scheduled headways.
Deviations from scheduled headways may be calculated as an absolute value or a
relative value. Absolute deviations are simply the difference between scheduled and
actual headways, expressing, for example, a 1-minute deviation for an observed
headway of 4 minutes with a scheduled headway of 5 minutes. Using this measure of
76
A Framework for Analyzing Transit Reliability
deviation, the center of the distribution is expected to be close to zero because the
average headways should be close to the scheduled headways.
Relative headway values are the expression of the observed headways based on the
scheduled headway. Thus, the same 4-minute observed headway can be expressed as
"0.80 the scheduled headway" (headway ratio value) or the headway deviation as "-0.20
the scheduled headway". For these values, the center of the distribution is expected to
be close to 1 if the relative observed headway is used, and 0 for relative deviations.
2. Wait Times
The expected wait time measure reflects the impacts of headway or schedule variability
on passengers. For low-frequency routes, passengers tend to time their arrival at a stop
with the bus' expected arrival to minimize their wait time. The expected wait time on
low-frequency routes is a function of on-time performance (schedule deviations) and the
and the proportion of passengers that time their arrival to schedules and passengers
that arrive at random (Bowman and Turnquist, 1981).
Expected wait time is a particularly good measure of service reliability for passengers on
high-frequency routes, where passengers are assumed to arrive at random without
consulting posted schedules. For perfectly even headways, the expected wait time is
simply half the headway. However, the expected wait time increases as the headway
variability increases, as described in Section 3.4.2.
W
2
= - 1+ cov2(h)]
2
Another measure of passenger wait time is the excess passengerwait time (EWT),
which measures the difference between actual expected wait time and the expected wait
time if headways were as scheduled. It is an appropriate measure to reflect the change
in average expected wait time due to poor headway adherence and unreliable service.
h
EWT = w - -[1+
2
cov 2 (hs)]
where w is the expected passenger wait time, h is the mean scheduled headway and
cov(h,) is the coefficient of variation of scheduled headway (Wilson et al. 1992).
Another measure of excess headway time can be formulated by calculating the percent
of excess time, ET, given as the sum of positive deviations over the total elapsed time.
77
Chapter 3
Z(x -y )
ET = j
for i e A,jeB
zxi
where x is the actual headway, y is the scheduled headway, A is the set of all
headways and B is the set of headways where the actual headway is greater than the
scheduled headway (note that B is a subset of A). For example, if scheduled
headways are 5-minutes and actual headways were 8, 4, 7, 3, 9, 2 minutes, then the
excess time is equal to 0.27. This can be interpreted as approximately 27% of the
elapsed time was spent in excess. It is another measure of longer passenger wait times
due to headway variability.
3. Overcrowding
Passenger loads are typically measured as the ratio of passenger load to the number of
available seats. Overcrowding is a measure of high passenger loads, where the number
of passengers exceeds a threshold value which affects passenger comfort levels. It also
compares actual crowding levels with service standards on the maximum number of
standees on each trip or averaged over a span of time. It is also related to reliability
because high load levels increase dwell times and possibly wait times (see Section
3.3.2).
4. Percent of Unreliable Trips
These measures describe the number of trips falling outside some service reliability
standard. The percent of unreliable trips is a useful measure to describe overall
performance and the probability of unreliable service. Transit agencies may have
service policies that promise service above a certain standard, such as "95% of trips
arriving within 0-5 minutes late". Thus, percent of unreliable trips is a measure of the
extent of poor service. The percent of unreliable trips is measured for a number of
service attributes, each with a given threshold value (or range) for what is considered a
reliable (or unreliable) trip.
78
-
Late departures. The percent of trips that leave the terminal (or time point) more
than x minutes later than schedule, with the value of x determined by the transit
provider.
-
Early departures. The percent of trips leaving the terminal (or time point) early by x
minutes from its scheduled departure time. The percent of early departures can be
combined with the percent of late departures to determine the percent of trips with
on-time departure from the given terminal (or time point).
-
Late arrivals. The percent of trips arriving at the terminal (or time point) more than x
minutes late. Because variability is known to occur in running times, this measure is
typically applied to the terminal to evaluate recovery times necessary to ensure an
on-time departure for the following trip. Another useful measure is the number of
A Framework for Analyzing Transit Reliability
trips arriving late by more than the scheduled recovery time. This indicates whether
running times and recovery times are sufficient to avoid late departures for
subsequent trips.
These measures are calculated for all trips each different time period, to account for
time-of-day variability. The next two measures can be calculated for all headways within
different time periods, where a headway is considered unreliable if it falls outside the
threshold value (or range) at any point along the route.
-
Bunches. Measured as the number of headways below a threshold value where the
buses are considered to be "bunched". The threshold value may be an absolute
value, such as 1 minute, or a relative value based on the scheduled headway, such
as the headway ratio or headway deviation.
-
Large Gaps. With the above measure, it is an indicator of headway adherence. This
measures the number of headways greater than a threshold value of what is
considered a "large gap" in service. Again, the threshold value may be an absolute
value, such as x minutes, or a relative value based on the scheduled headway, such
as the headway ratio or headway deviation.
Late gara-ge pull-out applies to specific work assignments that begin with a pull-out from
the garage, not individual trips. It is the same as the late departure measure except
taken at the garage, counting the number of work assignments that begin x minutes late.
This is a good measure to identify possible problems regarding operator behavior,
resource (driver or vehicle) availability or supervisory failures at the depot.
Late relief applies to operator duties that begin at a point in the route (street reliefs). It is
similar to late garage pullout in counting the number of work assignments that begin x
minutes late, and is an indicator of poor operator behavior or unrealistic scheduling.
D) Threshold Values
For all of the service measures described above, the transit provider must make
decisions on the threshold values or ranges of values to classify a service as reliable or
unreliable. These should be based on the level of service the transit provider can costeffectively deliver, as well as customer expectations, and may vary by type of route and
time of day.
As previously noted, most customers on low-frequency routes will time their arrival at the
stop based on the schedule modified by experience to reduce their expected wait time,
while passengers on high-frequency routes are assumed to arrive at random because
they place little confidence in the schedule but know their waiting time should be low.
Thus, the threshold values of deviations should be different for low-frequency and highfrequency routes due to the different impacts on passengers. The same applies to trips
across different times of the day with varying traffic conditions and passenger demands.
Variability in passenger demand affects how many passengers are being affected by
unreliable service. Variability in external conditions, especially during peak hours, may
79
Chapter 3
be so great that it affects the ability of the transit provider to maintain on-time
performance despite the best of schedule planning and service management.
The key points a transit provider should consider when making decisions on reliability
threshold values are discussed below in four groups: on-time performance, headway
adherence, passenger loads and percent of trips.
1. On-time performance
The transit provider has to define the range around the scheduled time within which the
bus is considered to be "on-time". This is necessary to give meaning to on-time
performance. These threshold values should depend on the type of route, scheduling
practices and location on the route.
For low-frequency routes, where a strictly scheduled-based perspective on on-time
performance is appropriate, schedule deviations affect passenger wait times. Early
departures may cause passengers to miss their scheduled trips and be forced to wait a
full headway for the next bus, while late departures simply increase the wait time of all
passengers waiting at the stop. Because of this, early departures should be avoided
and deviations in late arrivals kept to a minimum. The transit provider may implement a
"no early departure" policy and instruct operators to hold at time points until their
scheduled time, at the expense of increased travel times for passengers already
onboard. Thus, the transit provider may choose deviations between 0 to x minutes late
to be considered "on-time", with x determined by scheduling practices, travel
time variability, and customer expectations.
For high-frequency routes, the passenger concern is more on headway regularity, but
minimizing schedule deviations may still be preferred since headways will certainly be
balanced if schedules are maintained. The same criterion as for low-frequency routes
may be applied, but without being as concerned with early departures. The range for
"on-time" performance may then be given as (+/-) x minutes from the scheduled time,
with x again determined by scheduling practices, travel time variability and customers
expectations.
With regard to scheduling practices and travel time variability, the threshold value of x
minutes depends on how scheduled times at time points are set and managed.
Scheduled arrival times are typically set based on average running times, in some
agencies slack is built-in based on the running time variability at the end of the route in
the form of recovery time. If slack is built into the scheduled departure time at time
points, early buses should be held to avoid buses running early and the value of x
should be small. Otherwise, the value of x should account for the known variability in
actual arrival or departure times at the point. Thus, a trade-off between on-time
performance and travel times exists if the goal is to achieve tight on-time ranges.
Timepoint scheduling sacrifices travel times and speeds, as early buses are held
until their scheduled departure time.
80
A Framework for Analyzing Transit Reliability
There should also be a difference between the threshold values at the terminal and at
other points in the route. As described in Section 3.3.1, good on-time performance is
expected at the terminal because of recovery times and the higher degree of control at
these points. And because deviations tend to propagate down the route, transit
providers may want to be have tighter departure time ranges at the terminal.
The review summarized in Section 3.2.1 reveals that the most common current practice
is to set a 1 minute early to 5 minutes late range for "on-time" performance.
2. Headways
These values are used for high-frequency routes where the focus is on good headway
adherence. Transit providers must consider whether to use an absolute value or a
relative value to evaluate actual headways. Absolute values would be, for example,
considering any bus with a 1-minute headway to be bunched, while relative values would
consider a bunch any bus with an actual headway less than or equal to, for example,
0.30 of the scheduled headway.
Relative values may be preferred on routes on which consecutive headways are not
equal (example, they range between 3 to 5 minutes), so that appropriate comparisons
are made on actual and scheduled service and the true impacts on passengers are
evaluated. However, caution is given in using relative values. For example, a gap in
service might be considered as anything above 1.5 the scheduled headway. For service
with 3 minute headways, this gap in service would be a 4.5 minute headway, which
some passengers might still consider relatively frequent. But for service with 8 minute
headways, the impact on passenger time is an increase of 4 minutes (12 minute
headway).
3. Passenger Loads
Overcrowding threshold values are based upon seating capacity and an acceptable
maximum number of standees. It can vary by length of route, frequency of route and
time of day. For longer routes, the threshold value may be lower to ensure passengers
do not feel crowded and are more likely to have a seat for longer journeys. Transit
providers typically balance frequency with passenger loads. Low demand routes have
higher headways to avoid maintain productivity but with enough service to provide a seat
for all passengers, while high passenger demands justify the more frequent service,
where buses might be crowded but close enough to avoid long wait times. In peak
hours, threshold values for overcrowding tend to be set to higher values reflecting the
higher cost of providing peak hour service, due to spread penalties and 8-hr guarantee
work rules (Herzenberg, 1983).
Overcrowding threshold values may be set as the average maximum load over a time
period (or span of time, like 30 minutes), or in terms of number of trips with loads
exceeding a certain value. An example of the latter would be setting the threshold at no
more than X-percent of trips with passenger loads over Y-percent of the seating
capacity.
81
Chapter 3
4. Percent of Trips
Another common threshold is the percent of on-time trips that are needed to still
consider service quality as reliable. For each service attribute, this percent value may
vary depending on frequency of service and time of day. For low-frequency bus routes
and time periods, adherence standards may be lower given a good standard on
schedule deviations and the lower passenger demand. For example, good schedule
adherence may be stated as 75% of all trips with a schedule deviation of 0-5 minutes for
each time period. This acknowledges early departures are unacceptable and some
variability in departure times is inevitable. Deviations in schedule affect passenger wait
times, but do not impact operations as greatly as on high-frequency routes. For highfrequency routes and time periods, transit agencies may want to be more aggressive in
maintaining schedules or headways because of the higher levels of passenger demand
and stronger tendency for problems to propagate. Typical percent threshold values that
describe on-time performance and promised level of service follow those presented by
the Transit Capacity and Quality of Service Manual summarized in Table 3-1.
E) Performance Reports
Data recorded by automated collection systems are usually in formats that require both
pre- and post-processing. The raw data must be transformed into useful information that
operators and supervisors can use to analyze service reliability. Pre-processing
includes piecing together all of the data recorded by the vehicle, and it can either be
done automatically as the data is displayed for real-time tracking or when it is dumped
each night at the garage, or it may require human intervention. Post-processing involves
using these data items to calculate the service metrics previously described and
generate service performance reports. The outputs of this analysis are the calculated
service attributes that characterize service reliability for each trip, while performance
reports are the summary of these outputs and service measures that evaluate current
service for a given time period, and the extent of service unreliability.
Performance reports summarize the aggregate performance by time period accounting
for the trip-to-trip and daily variations in service delivery.
1. Running Time
The performance report should summarize the distribution of trip running times for each
time period, with the average, percentile and coefficient of variation (or standard
deviation) for comparison of observed and scheduled values. Actual running times
should be compared with scheduled running times, given by the current timetable, in
order to determine whether schedule adjustments are needed. Evaluations may include
calculating average (or percentile value) running time per time period to assess on-time
performance, or determining appropriate time period boundaries to construct a timetable
that reflects current conditions and the best possible values (minimizing variability) for
scheduled running time.
82
A Framework for Analyzing Transit Reliability
2. Schedule Deviations
Reports on schedule deviations evaluate on-time performance of buses focusing on its
distribution, along with the average and coefficient of variation by time period for
terminals and/or time points on the route. If scheduled running times are appropriate,
vehicles should closely follow the scheduled times, with small positive or negative
deviations, knowing variability is inevitable. Large schedule deviations indicate problems
such as inadequate running times, abnormal boardings, etc.
3. Headway Adherence
Performance reports on headway adherence include the distribution of headway
deviation or headway ratios, and the calculation of average and coefficient of variation
for comparison with scheduled values. Analysis also includes determining average
passenger wait times and evaluating the likelihood of extreme headway deviations, such
as bunching and large gaps. The evaluation of the likelihood of bunching and large
service gaps is important in characterizing service as highly unreliable with unpredictable
headways and with high probability of long wait times and crowding.
3.6.2 IDENTIFICATION
OF CAUSES
The identification of the causes of service unreliability, the second block of the analysis
process, involves a step-by-step approach to infer the underlying causes of the observed
reliability problems found in the previous section. It focuses on the observed deviations
from scheduled times and scheduled headways calculated using available trip data from
ADC systems.
The first steps in this process focus on deviations at terminals, examining the different
sources that may create such problems at the terminals and the effects of these
deviations on the rest of the trip. The deviations at the terminal may be the result of: 1)
deviation from previous trip, 2) operator behavior, or 3) supervision control. The later
steps analyze deviations at points down the route, where the sources of the problems
are different than at the terminal. Deviations developing along the route are likely to be
the result of: a) the trip beginning off-schedule initially and not being able to recover,
and/or b) a trigger somewhere between the terminal and the point of detection.
When analyzing a large set of data in order to identify the source of service reliability
problems, it is important to distinguish between random and systematic occurrences.
Random occurrences are hard to isolate and difficult to target because they are
unpredictable in terms of where and when they will occur, and their severity. Regardless
of how well designed or planned transit service is, random variations in demand, traffic
and operators will naturally cause fluctuations to arise. Many of these may have no
serious reliability impact. Systematic variations are easier to identify because of their
predictability, although this does not necessarily imply that they are any easier to
remedy.
83
Chapter 3
Random variations are due to inadvertent events which may never be repeated in the
same form. Their occurrence during the normal course of the daily operations does not
contribute to any pattern of unreliability. For example, the service reliability problems of
one morning peak period on one route may be attributed to a late departing bus. If this
bus trip is usually on-time and no other service problems are reported (normal traffic
conditions, no accidents, etc), the incidence may be isolated and the variation in service
for that day is considered random in nature. It might just be the operator had a hard time
starting the vehicle that day and was consequently late for his first trip. Delays due to
accidents on the road are typically considered random events in service.
On the other hand, systematic variations in service attributes are those which are
recurrent, for which a pattern of incidence can be identified. A single occurrence of
service unreliability can be considered as a systematic variation when it has similar
attributes of incidence, whether it is time of day or a causal trigger. For example, the
same reliability problem as above can be shown to be due to a single late departing bus
in the morning peak period. However, if performance reports show that the operator of
this particular late departing bus is consistently late on all his work duties, the cause of
unreliability may no longer a random variation, but a systematic problem caused by
inappropriate operator behavior. Similarly, if operators in general have a tendency to be
late on pullouts, this is systematic variation (caused by supervisory failures at the garage)
even if it affects different operators, or trips and routes are affected in different days.
This distinction is also important in targeting strategies to avoid them, or at least lessen
their impact. Systematic variations are typically somewhat predictable due to their
nature. The service variation in question has a high probability of occurrence because of
the similar factors involved. Random variations, on the other hand, are unexpected and
can only be dealt with after they have already triggered reliability problems. Thus,
systematic variations are usually better targeted by preventive strategies, and random
variations are better dealt with through corrective strategies.
In analyzing service reliability, even with large amounts of data, it is difficult to
characterize each and every deviation from schedule as random or systematic.
Indeed, there is a continuum of types of variations, ranging from completely random
to completely systematic, and focusing only on these extremes understates the
complexity of the problem. However, it is easier to determine whether there is a clear
pattern of occurrence. If there are a number of deviations which all have the same
root cause and share a common attribute, then the problem can be considered
systematic.
Attributes include, but are not limited to, time of day, day of the week, location, operator
and others. Thus, for example, the data may show that scheduled running times are
insufficient during the morning peak hour period, or that a particular operator is late in
reporting for 80% of his or her work duties.
The key is to determine a threshold beyond which to consider the overall problem as
systematic and of importance to merit the implementation of strategies to improve
service quality. If it is determined that X percent of deviations in a given time period
84
A Framework for Analyzing Transit Reliability
were the result of cause Y, then the decision must be made whether X is higher than
some operational standard to conclude the deviations are systematic and cause Y is
significant.
A. Deviations at Terminal
As described in Section 3.3.1, emphasis in this process is given to terminals because
they are the fundamental control points at which good schedule or headway adherence
is rooted. The terminal is a "restart" point, such that any deviations at the end of a trip
should not carry over to the next trip because of the scheduled recovery process. Also,
supervisors are frequently based at terminals to monitor and control departures to
ensure schedule (or headway) discipline. Thus, one expects very good on-time
departure performance at terminals.
Care is needed to ensure accuracy of time-at-location data at the terminals because
problems often exist in determining terminal times and passenger activities. Automatic
Vehicle Location (AVL) systems tend to record the arrival and departure at a time point
as the bus enters an area or zone (around a set of GPS coordinates). Typically, two
time points are defined at the terminal, inbound and outbound, which may create an
overlap in the catchment areas. This, as well as the fact that buses spend more time at
this time point (passenger boarding/alighting and/or recovery time), may create problems
if the system cannot accurately define (and distinguish) the arrival and departure time of
the inbound and outbound trips.
The focus is first on evaluating performance at the terminal and describing the service
reliability evolution process in terms of the effects of terminal deviations on the rest of
the route. Although some buses may be able to make up for an early or late trip start,
chances are these initial deviations will propagate further down the route and increase
unreliability. Then the focus is on describing a process to examine the causes of
deviations at the terminal. Terminal deviations from the schedule (time or headway)
may be the result of insufficient recovery time to remedy previous deviations, or human
factors related to operator behavior or supervision.
1. Performance and Effects of Deviations at the Terminals
The process to analyze the effects of deviations at the terminal depends on the
frequency of the route, with scheduled times the focus for low-frequency routes and
headways for high-frequency routes.
Schedule deviations. For low-frequency routes, the analysis is focused on schedule
adherence to investigate how departure behavior at the terminal affects performance at
points down the route.
1.
The schedule deviation is calculated at the terminal and specific points along the
route for each scheduled trip. The trips are categorized based on the deviation at
the terminal.
85
Chapter 3
The average and standard deviation of schedule deviations of trips within each
terminal deviation category are calculated for each specific point. This should provide
insight into the propagation of deviations along the route. Table 3-7 is an example of
the schedule deviation calculations at different points on the route. This analysis
shows how schedule deviations propagate along the route, separating trips by their
initial deviation at the terminal. This also helps identify points on the route where
deviations are the worse (or more variable).
Table 3-7. Example of Analysis of the Propagation of Deviations
Point A
Number
.
of trips
Deviation at Terminal
23
52
120
62
30
-2 to -1 minutes
-1 to 0 minutes
0 to I minute
1 to 2 minutes
2 or more minutes
Mean
schedule
Standard deviation
of schedule
deviation
-1.2
-0.5
0.2
1.3
3.4
deviations
3.2
2.1
1.2
2.3
2.4
deviation
Point B
Standard deviation
of schedule
deviations
3.5
2.2
1.5
2.6
2.8
-0.8
0.1
0.8
1.6
3.8
2. To further analyze the impacts of deviations at the terminal, all trips in each terminal
deviation category can be categorized according to the deviation at a specific time
point (call this point B). The result is an n by n matrix, such as the example shown in
Table 3-8, of the number of trips with a given combination of deviation at the terminal
point and time point B.
Table 3-8. Time Point Schedule Deviations vs. Terminal Deviation
Schedule Deviation at the Terminal
At Time
Point B
< -2 mins
-2 <= x < -1
-1 <=x<O
0<=x<1
1 <=x<2
2<=x<3
>= 3 mins
< -2 mins
-2<= x<-1
-1 <=x<0
0<=x< 1
1<=x<2
2<=x<3
>= 3 mins
7
3
2
0
2
0
0
5
1
1
0
0
1
0
12
4
7
3
0
0
0
53
49
71
48
29
11
3
4
11
36
34
26
16
13
0
3
6
8
13
10
12
1
0
0
6
8
11
49
From the example above, the following observations can be made. Variability in
deviations at time point B are significant given on-time departures at the terminal, if
an on-time departure is considered a departure between 0 and 2 minutes after the
scheduled time. Early departures seem to remain ahead of schedule down the
route, while late departures have more variability but tend to remain behind
schedule.
Using the transit provider's range of values for "on-time performance", one can also
compare the number of trips that began "late", with those at points further
86
A Framework for Analyzing Transit Reliability
downstream. The distribution of trips would give a good indication of how many trips
are able to make up for given initial deviations.
Trips that have been reassigned or that are out of the scheduled order must be
discarded to avoid bias, unless adjustments in the scheduled-time data have been
made to reflect the new scheduled trip.
Probability of delay. Schedule deviations can also be used to analyze the probability of
delays at a specific time point given a deviation at the terminal.
1.
For all trips with on-time departure at the terminal, the schedule deviation is
calculated for both the terminal and the specific time point (call it point B).
2. The schedule deviation distribution can be computed, making sure there are a
sufficient number of trips to support the level of data aggregation. From this
distribution, the probabilities of on-time arrival at point B can be calculated.
3. The same distributions can be generated for each terminal departure deviation
category. Then, the probabilities of deviations at point B are compared given the
departure behavior at the terminal. The expectation is that the greater the positive
(or negative) deviation at the terminal, the higher the probability of positive deviation
at point B.
4. The distribution of trips with on-time departures also gives insight into the adequacy
of the scheduled running time between the terminal and point B. If the probability of
arriving at point B within an "on-time" range is low, it is a good indication that the
scheduled times are unrealistic and need adjustment.
Headway Adherence. For high-frequency routes, the same type of deviation analysis
can be made using the headway deviation values instead of the schedule deviation
values. Instead of schedule deviation at time points, the headway ratio value is used to
correlate departure behavior at terminals with problems at points down the route. The
trips are categorized by their initial headway ratio.
1.
The headway ratio (scheduled over actual headway) at the terminal and specific time
points are calculated for each scheduled trip with valid headway data.
2. The average headway ratio and standard deviation of all trips are calculated for each
category and at each time point. This provides a picture of whether vehicles
maintain their initial headway throughout the route, and what the tendency is in terms
of headway adherence problems. Table 3-9 is an example table of headway ratio
calculations based on a trip's headway ratio at the terminal.
87
Chapter 3
Table 3-9. Example of Headway Adherence Analysis
Headway Ratio at
terminal
Total #
trips
of
HR < 0.3
0.35 HR < 0.5
0.55 HR < 0.7
0.7: HR < 1.3
1.3:5 HR < 1.5
HR 2 1.5
20
36
35
120
20
25
At oint A
Std. Dev.
Average
HR
of HR
0.15
0.45
0.65
1.01
1.24
2.0
At point B
Std. Dev.
Average
HR
of HR
0.23
0.15
0.26
0.24
0.26
0.57
0.10
0.41
0.58
1.02
1.35
2.3
0.20
0.12
0.23
0.24
0.23
0.52
3. Again, all trips in a given category are categorized once more according to the
headway ratio at a specified time point (point B). A similar n by n matrix, such as the
example provided in Table 3-10, summarizes the number of trips with a combination
of its headway ratio at the terminal and at the given time point. This matrix provides
some insight into the tendency of headway deviations to propagate and create
bunches and gaps in service.
Table 3-10. Example of Headway Ratios per Terminal Headway Ratios
Headway Ratio at
Ratio at Terminal
_Headway
Time Point B
0-0.5
0.5-0.75
0.75-1.00
1.0-1.25
1.25-1.5
>= 1.5
0-0.5
0.5-0.75
0.75-1.00
1.0-1.25
32
8
2
2
21
22
9
4
38
21
37
20
24
18
19
22
5
2
10
7
2
3
4
7
1.25- 1.5
0
1
12
26
11
12
>= 1.5
1
2
14
37
22
34
This example shows high variability in headways at time point B even with good
headway departure from the terminal. It also reveals that buses that depart bunched
(0.5 or less of the scheduled headway) have a strong tendency to remain bunched,
and those that depart with a large gap in service (1.5 or more) have a strong
tendency to remain spaced apart.
2. Causes of Deviations at the Terminal
The causes of deviations at the terminal are evaluated focusing on the recovery time
between trips. Recovery time is the time at the terminal between the end of a bus'
previous trip and the beginning of its next trip. Recovery time is intended to ensure that
buses are able to depart on schedule despite any deviations in its previous trip, as well
as to give operators a break between trips.
Under normal operating procedures, operators are expected to use the recovery time to
avoid any deviations from carrying over. However, if recovery time is not enough for
buses to be able to start their next trip on-time or for operators to take a needed break,
then the next trip will depart late from the terminal.
88
A Framework for Analyzing Transit Reliability
The scheduled, available and actual recovery times, as well as the scheduled and actual
headways, are calculated for each trip and compared to determine the root cause of
deviations in departure time.
Scheduled recovery time is the amount of scheduled time at the terminal between
consecutive trips of a vehicle. Calculated as the time between the scheduled arrival time
of the previous trip and the scheduled departure time of its subsequent trip, it is often
determined by the variability in running times, minimum operating policies, or clock-face
scheduling practices.
Available recovery time is the time between the actual arrival time at the terminal and the
scheduled departure time of its subsequent trip. It is the time a bus has at the terminal if
it were to depart its next trip on-time.
Actual recovery time is the time between the actual arrival time at the terminal and the
actual departure time on its next trip. It is the time the bus actually spent at the terminal.
The difference between the actual recovery time and the available recovery time is the
schedule deviation in departure.
When analyzing the available recovery time and actual departure times at the terminal, it
is important to consider the time needed for passengers to board and alight. It is
unreasonable to expect an on-time departure for a bus that arrives with an available
recovery time of 1 minute at a busy terminal, because there will likely be passengers in
the bus who will be alighting and passengers at the terminal who are waiting to board.
This terminal passenger processing time will depend on passenger demand as well as
terminal configuration, and it will vary across routes and across transit agencies.
One approach to determine a reasonable range of time buses typically spend at the
terminal boarding and alighting passengers is to look at the departure behavior of trips.
An m by n table of the number of trips with a given combination of available recovery
time and schedule deviation shows the variability of departure deviations based on the
available recovery time. With a large enough sample to offset any bias from operator
behavior, an estimate can be made of the time buses must spend at the terminal for
passenger boarding and alighting. An example is presented in Table 3-11, where it is
inferred that most buses take 2 to 3 minutes at the terminal. This estimate is based on
the observations that buses with 1 to 2 minutes of available recovery time tend to depart
the terminal I to 3 minutes late, and buses with an available recovery time of 2 to 3
minutes tend to leave 0 to 2 minutes late.
This analysis may also provide insight into the scope of problems at the terminal. The
example below shows a large variability in departure deviations for trips with enough
recovery time. For buses with 5 or more minutes of recovery time, one would expect a
good on-time departure performance. However, deviations vary quite a bit, with some
buses leaving very early, while others are very late. This indicates that there are other
causes at the terminal triggering poor departure performance. The cause of such
deviations may be operator behavior or control interventions by a supervisor.
89
-77-777-7
MRMMW"TTM
Chapter 3
Table 3-11. Terminal Departure Schedule Deviation vs. Available Recovery Time
Available Recovery Time
Departure
Schedule
Deviation
<=-2
mins
-2to-1
mins
-1to0
mins
Oto1
mins
1to2
mins
2to3
mins
3to4
mins
4to5
mins
5+
mins
0
8
4+ mins early
3-4 mins early
2-3 mins early
1-2 mins early
0-1 mins early
0-1 mins late
1-2 mins late
2-3 mins late
-
-
-
-
-
-
-
0
-
-
-
-
-
0
0
-
-
-
-
0
5
-
-
0
-
2
0
3
1
3
2
0
0
3
7
2
5
12
8
3
14
6
1
1
4
19
11
8
16
121
52
15
2
3
5
39
115
40
8
3-4 mins late
1
1
4
4
1
2
2
7
4
4+ mins late
28
7
6
4
4
2
1
6
3
0
6
By comparing the three values of recovery time (scheduled, available and actual),
inference on the causes of deviations at the terminal may be drawn. If a significant
number of buses do not have enough available recovery time because of persistent late
arrivals, then the late departures might be attributed to the current scheduled times and
adjustments to the timetable may be needed. On the other hand, if most trips have an
available recovery time (more than the determined minimum to account for variability in
late arrivals, passenger demands and operator breaks), then deviations in departure
may be due to operator behavior or control intervention.
However, this analysis does not account for the relationship between schedules and
operator behavior. Schedule recovery times also serve as breaks for operators. If
schedules are tight and available recovery times are typically lower than scheduled,
operators will not be getting the amount of break time they are entitled to. This leads to
low morale and added stress on operators, who may tend to take their scheduled
breaks, regardless of the available recovery time and departure deviation. And while
this my be viewed as operator behavior, it may be exacerbated by poor scheduling.
The following steps are proposed to make inferences about the cause of departure
deviations at terminals, based on the recovery time, the schedule and the headways.
1.
The scheduled, available and actual recovery times are calculated for all trips (with
valid data) at the terminal. Determine the "minimum" recovery time, as described
above, to infer the amount of time buses typically spend (and need) at the
terminal for passengers to board and alight.
2. Late arrival. An available recovery time less than zero indicates that the trip could
not depart on-time because of a late arrival. However, it is useful to examine the
amount of actual recovery time that was taken. Beyond the time needed for
passengers to board and alight, one would expect the bus to depart as soon as
possible, given that it is already behind schedule. Thus, even though it is the late
arrival that causes the late departure from the terminal, operator behavior may also
90
A Framework for Analyzing Transit Reliability
contribute by extending this time and causing further delay. If a large number of trips
are arriving late and do not have enough recovery time, schedules need to be
adjusted. Operator behavior may also be a result of tight schedules, if operators
believe they deserve a break but are not getting it because they have to depart for
their next trip almost immediately.
3. With recovery time. For other trips that have some recovery time at the terminal, the
approach to infer the causes of unreliability depends on the route frequency (lowfrequency and high-frequency routes) and whether there is a supervisor stationed at
the terminal. Terminal supervisors are responsible to ensure good on-time
performance or manage departures to balance headways.
A. For low-frequency routes, the focus is on on-time performance and schedule
deviations. Buses ought to strictly follow schedules and headway-based control
actions such as holding are inappropriate. In this case, deviations can be
attributed to poor supervision and operator behavior. Poor supervision is
responsible for poor on-time performance if there is a terminal supervisor. If
there is no supervisor at the terminal to enforce on-time departures, then any
schedule deviation beyond the acceptable on-time range of a trip with available
recovery time, is the result of operator behavior.
B. For high-frequency routes, maintaining headways might be a more realistic
objective than keeping to schedules and departures from terminals may be
affected by headway-based control actions. Thus, schedule deviations at the
terminal may be the result of headway-based control actions, and not entirely the
result of operator behavior. In this case, there are four cases considered below.
Documentation of control actions by terminal supervisors would help support the
inference of causes of departure deviations for cases (2) and (3). This can be
information on implemented control actions or a log of communications between
supervisors and operators. If no information is available, a number of
assumptions will have to be made regarding operator behavior and departure
deviations at terminals.
a. An early departure with a smaller (than scheduled) preceding headway. It is
likely that no control actions were implemented here because the headway
would have been closer to scheduled if the bus had departed on-time. Thus,
the early departure can be attributed to operator behavior or poor supervisor
performance.
b. An early departure with a preceding headway greater than or equal to the
scheduled headway. In this situation, the previous bus must also have left
earlier than scheduled or that trip had been missed. There is the possibility
that, if present, the supervisor instructed the operator to depart early in order
to maintain headways. Otherwise, the early departure is attributed to
operator behavior.
91
Chapter 3
c. A late departure with a preceding headway less than or equal to the
scheduled headway. In this case, the previous bus must have left the
terminal very late and it is possible that the supervisor instructed the operator
to hold at the terminal and depart late to maintain more regular headways.
d. A late departure with a large preceding headway. An on-time departure
would have resulted in a more balanced headway. Thus, it is unlikely that
holding was implemented as a control action, and the late departure can be
attributed to operator behavior.
An example of this analysis on a high-frequency route is presented in Table 3-12.
The trips analyzed have acceptable values of available recovery time. The
example shows that many of the trips with a high departure headway also had a
late departure, and most of the trips with a low headway departed before their
scheduled time. For these trips, an on-time departure would have resulted on
headways closer to schedule (ratio closer to 1). Therefore, these deviations are
inferred to be caused by poor operator behavior because any control actions by
terminal supervisors would have improved the departure performance (holding to
balance headways or until scheduled). If it is known that the transit provider
does not usually implement holding strategies at the terminal to maintain regular
headways, then operator behavior can be inferred as the cause of the departure
deviation.
Table 3-12. Terminal Departure Deviation vs. Terminal Headway
Terminal Departure
Deviation
5 or more minutes late
1 or more minutes early
Ratio > 1.5
HeadwRatio < 0.5
(total number of trips = 100)
(total number of trips = 84)
45
7
12
60
A trend of early and late departures by an operator might give insight into the
likelihood that the departure deviation in cases (2) and (3) was the result of
operator behavior and not a holding strategy. The analysis helps infer the
tendency of an operator to have poor terminal performance. It is unlikely that an
operator is frequently going to be instructed to depart off-schedule or have
extreme delays at the terminal due to passenger demand or traffic that prevent
good on-time departure performance.
Missed trips present a problem in this analysis. If missed trips are not
specifically marked in the data records, then it cannot be determined whether
missing data is due to missed trips or errors in AVL recording. It is difficult to
determine departure behavior from those trips with a preceding bus with missing
data. One way to address this is to discard headway records with unknown
headways, to avoid bias towards large headways that cannot be known to be
missed trips or "ghost" buses.
92
A Framework for Analyzing Transit Reliability
Deviations at Other Points
For time points other than the terminal, a different analysis process is developed. It is
applied to trips that fall out of the "on-time" window, determined by the transit provider's
threshold values. Illustrated in Figure 3-9, the analysis consists of the following
sequential steps to infer the causes of deviations:
1.
The first step is to determine whether there was a deviation at the terminal to cause
the bus to begin its trip off-schedule. Because deviations tend to propagate along
the route, deviations at other time points are likely to result from a departure
deviation at the terminal. The terminal deviation is marked as the cause of
unreliability for any trip with deviations down the route that began with an "offschedule" terminal departure. For low-frequency routes, scheduled times at the time
point and terminal are used, while for high-frequency routes, the focus is on the initial
headway at the terminal.
2. If the bus departed the terminal on-time, a significant deviation from the norm in
passenger loads somewhere on the route may cause the bus to deviate from
schedule. Passenger counts and dwell time analysis help determine whether an
abnormally high count of passenger boardings or alightings caused the bus to fall
behind schedule, or low demand (not having to serve all stops) caused the bus to run
ahead of schedule.
3. Schedule or headway deviations may be the result of inadequate scheduled times at
time points. Previous running time and schedule deviations evaluations, as describe
in Section 3.6.1, may show current scheduled times need adjustments, and
therefore, such deviations can be attributed to problems with the timetable.
4. Operator behavior. If the bus had an on-time departure and current scheduled times
are acceptable, then schedule deviations at time points may be due to operator
behavior. The analysis is to look for operators that tend to deviate from schedules
(tendency to be running early or late, or catch up to its leader and create a bunch).
5. Externalities. Event-records and speed profiles (level of detail D or E, according to
Furth et al. 2003) may provide insight into traffic conditions to help make
assumptions on the cause of the schedule deviations. For example, if the speed
profile shows constant start and stop motions, high traffic volume may be the cause
of service delays.
93
Chapter 3
Figure 3-9. Sequential Process to Identify Causes of Deviations
Schedule or
headway
deviation
Bus began trip offschedule, deviation
propagated down the
route
Deviation at
terminal?
No
Yes-+
passenger
Ii1itin
_
Off-chedule because
of high or low
passenger demand
_
_
_
_
_
_
_
_
No
e schedul
inadequate
for
na
-Yes--*i
or
Operator behavior
Unreliability caused by
schedule planning
+poor
Extemalities
The end-product of the analysis is a summary report on service reliability problems and
the attributed causes of such problems based on the analysis findings. The goal is to
provide a bigger picture of where major problem spots are in the route and what are the
main causes of service unreliability in order to select effective preventive and corrective
strategies to improve overall service.
3.6.3 APPLICATION OF STRATEGIES
The last block of analysis in the proposed reliability analysis process is the evaluation
and application of strategies to improve service reliability. As described in Section 3.4,
there are a number of potential strategies to improve service reliability, targeted at
reducing the occurrence of reliability problems or lessening the impacts on service and
passengers once service has deteriorated.
The analysis will show where and when problems have a tendency to develop, and
which are the triggers or causes that have the greatest impacts on service reliability.
Knowing what the causes of unreliability are on a route helps the transit provider's
94
A Framework for Analyzing Transit Reliability
decision-making process on what strategies are the most applicable to current service
and most practical to implement.
This part of the analysis involves applying the theoretical relationships between causes
and strategies described in Section 3.5 to the practical analysis results from Sections
3.6.1 and 3.6.2. Figure 3-10 illustrates the application of potential strategies based on
the process to identify the causes of unreliability problems. A summary of the most
applicable strategies are:
-
Terminal procedures. Recovery times at the terminal are key to ensure trips begin
on-time and do not create reliability problems that tend to propagate down the route.
Good service management is needed by way of supervisors at terminal to ensure ontime departures or good headway control, and operator training helps operators
understand the importance of following schedules and imparting better judgment on
managing reliability.
-
Schedule planning. If the cause of schedule deviations is unrealistic schedules that
buses are not able to maintain, then schedule planning is needed to adjust the
running times, scheduled times and headways to better reflect actual service and
adjust for variability.
-
Training and policies. This applies to both operators and supervisors. Variability due
to operator behavior can be reduced by better training practices or stricter
enforcement of policies regarding on-time departures and service performance.
Penalties also apply to supervisors, especially at terminals, with poor performance in
service management who do not enforce schedules or implement effective control
strategies to reduce reliability problems.
-
Corrective strategies. Variability in externalities and passenger demands, beyond
those that can be accounted for in schedule planning, are unavoidable. These are
better targeted through corrective strategies and service management to restore
service back to normal as quickly as possible.
95
Chapter 3
Figure 3-10. Application of Strategies
Schedule or
headway
devlation
Aopoation ol Strate-aies,
-
Bus began trip offschedule, deviation
propagated down the
route
Deviation at
terminal?
Adjust recovery times
Better supervision to manage
schedules and headways
Reduce variability from
operator behavior (training,
penalties, etc).
I
I
I
No)
noe
passenger
ingf
-
Of-schedule because
of high or low
-
Unreliability caused by
poor schedule planning
Control strategies: express,,
speed-uptslow-down
passenger demand
No
schadulI
inadequate f
y
Y
ndaer
-
N
Operator behavior
Adjust scheduled running times
Adjust scheduled times at lime
points
No_
Externalities
1~
I
I.
*
1-v.
Reduce variability through
exclusive lanes or signal priority
Corrective strategies: Better
service management in case of
accidents, breakdowns, hightraffic volumes or delays due to
weather
Operator training
Penalties and dlsclplinawy
actions for poor service
Control strategies to reduce
variability from speed
differentials
Transit providers must also assess the practicality and cost-effectiveness of potential
strategies. There are some strategies that can easily be implemented and tested for
effectiveness. Implementation of strategies such as changes in the schedule, on-time
enforcements policies (no early departures, holding at terminal, etc), training, or
corrective control-actions, can be assessed through before and after studies to analyze
the resulting changes in service performance. There are capital costs involved with
operator training and supervision, but these are not permanent changes and can be
implemented as a test before full application. The cost of performing these types of
analysis is low because of the availability of large data sets from automated data
96
A Framework for Analyzing Transit Reliability
collection systems. However, other strategies, such as exclusive lanes or signal priority,
must be assessed before implementation. The high-cost of application and large-scale
changes, does not allow these preventive strategies to be put into practice and then
measured for effectiveness. These strategies are better assessed through predictive
models.
Predictive Models
The identified causes and potential strategies can be analyzed further by developing and
applying regression analyses and simulation models. This framework analysis serves as
a guide to understanding service reliability and identifying the initial triggers of service
reliability problems, and proposes the use of computer models to test the significance of
the identified causes of service unreliability and to evaluate the effectiveness of specific
strategies.
The large amounts of data available through automated data collection (ADC) systems
allow for detailed analysis, such as those developed by Strathman et al. (2001, 2003), to
provide more insight into the true impacts of each of the different causes on service
variability. These regression models can contribute to the proposed framework analysis
by determining more operator specific factors that affect running times and departure
deviations, or evaluating the effects of deviations at the terminal on points further down
the route. One useful model would be to estimate schedule deviation as a function of
the initial deviation at the terminal point, time period, operator-specific characteristics,
headway, passenger demands, and other variables.
Data from ADC systems can also be used to develop simulation models to analyze
service reliability, test the significance of the identified causes, and evaluate the
proposed strategies according the identified sources of problems. A simulation model
using AVL and APC data from the Chicago Transit Authority was developed by Moses
(2005) with the intent to evaluate the benefits of control strategies on a route. Moses
was not able to calibrate and validate the simulation model, which showed significant
differences between the simulated and actual service statistics of headway regularity
and travel times. The simulation showed a stronger propagation of headway irregularity
(bunching) than actual service, especially in the second half of the routes tested. The
author concludes that this may be the result of operator behavior and service manager
decisions to maintain headway regularity and schedule adherence, which are not
included in the model. Perhaps, the simulator can be improved by the implementation
and results from regression models on operator behavior, headway regularity and
passenger loads, such as those by Strathman et al. (20011, 2003).
While the simulation model (Moses 2005) could not be validated, the development of
such simulation models using automated data collection systems remains a good
approach to apply and test the characteristics of service reliability outlined in this
research.
97
Chapter 3
3.7 SUMMARY
This chapter reviewed the most significant causes of unreliability, potential strategies to
improve service reliability, and the complex interrelationships between them. A practical
framework is developed to evaluate service performance and summarize service
reliability. First, service reliability is characterized using data available from Automated
Data Collection (ADC) systems and appropriate service measures, such as the
distribution of running times and headways. Then, service performance is evaluated to
infer the causes of unreliability. The framework first looks into performance at the
terminal and then at other points on the route, following a number of sequential steps to
infer the causes of unreliability. At the terminal, deviations may be caused by
inadequate schedules (recovery time does not account for running time variability), poor
operator behavior and poor terminal supervision. Causes of deviations at other points
on the route include initial deviations at the terminal, passenger loads, unrealistic
schedule times, operator behavior and externalities. Based on the results of these
analyses, transit providers can evaluate what strategies are the most applicable to
current service and most practical to implement.
The next chapter presents an application of the framework analysis using archived data
from the Massachusetts Bay Transportation Authority (MBTA) Silver Line Washington
Street route.
98
4.
CASE STUDY:
APPLICATION TO MBTA
The previous chapter outlined a process to analyze service performance and assess
service reliability of a bus route. The chapter reviewed the most common causes of
unreliability, their complex interrelationships and strategies to deal with them. It also
proposed a process to use archived data collected from automated systems to analyze
the extent of service unreliability on a bus route, identify the specific causes of problems
and to guide transit providers in identifying effective strategies to improve service quality.
In this chapter, the proposed reliability analysis process developed in Chapter 3 is
applied to a Bus Rapid Transit (BRT) route at the Massachusetts Bay Transportation
Authority (MBTA) system: the Silver Line on Washington Street. The purpose is to
examine service reliability on this route, and recommend improvement strategies based
on the identified causes of the reliability problems. Section 4.1 presents a brief
description of the case study route, while Section 4.2 describes the route's available
archived data and Section 4.3 presents the scope of analysis and data assumptions.
The results of the service reliability assessment analyses are presented in Section 4.4.
Section 4.5 presents the results of the analyses to identify the causes of unreliability,
while the strategies to improve service reliability on this route are presented in Section
4.6. A summary of the findings is discussed in Section 4.7.
4.1
MBTA SILVER LINE
ON WASHINGTON STREET
The Massachusetts Bay Transportation Authority (MBTA) is the oldest and one of the
largest public transportation systems in the United States. The MBTA system serves
175 cities and towns, and comprises rapid transit, bus rapid transit, bus, commuter rail,
water ferry, contracted bus and paratransit services. Ridership is over 1.1 million
passenger boardings a day, with approximately 792,600 one-way passenger trips per
day. Bus and trackless trolley service involves 162 routes and 9,000 stations/stops, and
carries around 30% of the average daily passenger boardings (MBTA website, 2005).
The Silver Line is Boston's first Bus Rapid Transit service, and is currently being
implemented in three phases. The Washington Street corridor is the first phase, which
opened in June 2002, providing service between Dudley Square in the south-west of
Boston, and the Downtown area. The second phase, the Waterfront line, began service
in December 2004, and offers service between South Station (Red Line subway station)
and South Boston, and to Boston's Logan International Airport. The final phase is
currently in planning and design and is expected to connect the first two phases with a
tunnel through the downtown area.
A Bus Rapid Transit (BRT) system is "a flexible, rubber-tired form of rapid transit that
combines stations, vehicles, services, running ways and ITS elements into an integrated
system with a strong identity" (Levinson et al. 2003. The main objective of a BRT
system is to provide convenient, fast, frequent and high-quality bus service. Bus Rapid
99
Chapter 4
Transit is typically characterized by dedicated lanes, distinctive stations and vehicles,
off-vehicle fare collection, Intelligent Transportation Systems (ITS) technologies, and
frequent service.
This case study focuses on the Silver Line on Washington Street because, as the
MBTA's first BRT system and flagship bus service, it should provide exceptional service
reliability. It is considered Boston's fifth rapid transit line, and major capital investments
have been made to offer high-quality bus service on this corridor. The dedicated lanes
allow for faster travel speeds and reduced travel times, state-of-the-art vehicles and
stations provide higher levels of comfort and convenience to passengers, and the
implementation of ITS technologies allow for better operations control and management.
The Silver Line on Washington Street (see Figure 4-1) is approximately 2.3 miles in each
direction and serves around 14,100 weekday passenger trips (MBTA website, FY04).
Connections to three of Boston's four rapid transit lines are available at the New England
Medical Center (Orange Line), Downtown Crossing (Red Line and Orange Line) and
Boylston (Green Line) subway stations.
100
Case Study: Application to MBTA
Figure 4-1. Silver Line Washington Street Route
Downtown
rossIng
Temsple
South
Station
HynesCop
vention-
edi
Suth End
Prudental,
it
East
Berke'
Symphony'
/
Massachusetts
Avenue
dway
nion Park
awtoti St
T Ruggle
Worc ister Square
Massach Lisetts Aven9
/,)Lenox St
-MeInea Cass BIwd
Dudley Square
Service begins at the Dudley Square station, and runs inbound along Washington Street
towards Downtown Boston, ending at Temple Place, close to the heart of the Boston
commercial area at Downtown Crossing and Park Street. Outbound service begins at
Temple Place, operating on Boylston Street, which forms a one-way pair with
Washington Street in the downtown area, before returning to Washington Street and
back to Dudley Square.
The buses operate on a non-protected exclusive bus lane for most of Washington Street
and in mixed-traffic going through the Downtown area. Non-protected exclusive bus
lane means that bus travel is still affected by turning movements at intersections, illegal
travel and double-parked vehicles. The route is also equipped with signal priority
technology; however, it is still in its testing phase and currently not being used. Each
bus stop is sheltered, and provides seating and bike rack amenities. They are also
equipped with a "smart kiosk" that provide passengers with schedule information,
101
Chapter 4
variable message signs, area maps, and a police call box for emergencies. Table 4-1
presents the 12 bus stops in each direction and the approximate spacing between them.
Table 4-1. Silver Line Route Stops
Stop Name
Inbound
Outbound
Dudley Square
Approximate
Distance (miles)
---
Melnea Cass Blvd.
0.25
Lenox St.
0.13
Massachusetts Ave.
0.12
Worcester Square
0.12
Newton St.
0.12
Union Park St.
0.30
East Berkeley St.
0.25
Herald St.
0.13
New England Medical Center
0.30
Chinatown
0.20
Temple St.
0.20
Temple St.
--
Boylston St.
0.25
New England Medical Center
0.35
Herald St.
0.30
East Berkeley St.
0.15
Union Park St.
0.25
Newton St.
0.30
Worcester Square
0.12
Massachusetts Ave.
0.12
Lenox St.
0.12
Melnea Cass Blvd.
0.13
Dudley Square
0.15
The Silver Line on Washington Street has two terminal points: Dudley Square and
Temple Place. Dudley Square is a major transfer station with high passenger activity
from 7 other connecting bus lines, and station configuration allows for multiple Silver
Line buses to be idle at this stop. Temple Place is also a major transfer point, located in
the downtown area with nearby access to the Downtown Crossing and Park Street
subway stations, and other connecting bus routes. It is located on a narrow, mixedtraffic one-way street, which makes it difficult to hold multiple buses at this terminal
station.
Recovery time at terminals allow for buses to recover from any previous delays and
begin their next trip on-time, and also give operators a short break between trips.
Ideally, the two terminals at the end of each direction would serve as a control point to
ensure on-time performance or good headway control for both inbound and outbound
trips. However, this route has an unusual two-terminal configuration that does not allow
102
Case Study: Application to MBTA
this type of recovery capability to be fully implemented. The physical constraints at
Temple Place limit the ability of buses to be held at this terminal point and the amount of
recovery time that can be scheduled. Thus, recovery times at Dudley Square are
greater, ranging from 3 to 11 minutes, while recovery times at Temple Place only range
from 1 to 4 minutes.
Running times vary throughout the day, but are typically around 20 minutes in each
direction. Table 4-2 shows the scheduled times of a trip in the AM peak hour period at
the terminals and at three intermediate points. For this sample trip, the inbound running
time is 17 minutes with 1 minute of recovery time at Temple Place. The outbound
running time is 14 minutes with 5 minutes of recovery time at Dudley Square Station
before the vehicle's next trip.
Table 4-2. Scheduled Departure Times of Sample Trip
Inbound Service (Dudley Square to Temple Place)
Dudley Square
Washington & East
Washington & East
Berkeley Sts.
Newton Sts.
Station
7:55
8:02
New England
Medical Center
Temp___Pac
8:09
8:12
Washington & West
Newton Sts.
DudeyStaton
8:22
8:27
1
8:07
Outbound Service (Temple Place to Dudley Square)
TpPa
New England
Medical Center
Washington & East
Berkeley Sts.
8:16
8:19
8:13
1
Headways also vary throughout the day, with a maximum of 15 minutes in the early
morning and late night time periods, and a minimum of 3 to 5 minutes in the peak hours.
Table 4-3 shows the scheduled headways on the route, with the scheduled departure
times of the first and last trips leaving Dudley Square and Temple Place.
Table 4-3. Silver Line Washington Street Route Schedule
Rush Hour
Service
First Trip
Dudley
5:15 am
Square
I
Temple Place
5 mins
_
5:31 am
Midday
Service
8 mins
I
5 mins
Evening
Service
10 mins
_
8 mins
Late Night
Service
12 mins
I
10 mins
Last Trip
LastTrip
12:43am
_
12 mins
1:00am
Variability in scheduled headways makes this route both a high-frequency and lowfrequency route. It is considered a high-frequency route in the rush hour and midday
service time periods, where scheduled headways are less than 10 minutes. It is
considered a low-frequency route during the early morning, evening and late night
service time periods, when scheduled headways are 10 minutes or more.
103
_
Chapter 4
Passenger demand typically follows the distribution illustrated in Figure 4-2, with an
average maximum load of 45 passengers in the peak hours8 . The inbound direction
carries a higher volume of passengers in the AM peak and first half of the midday period,
while the outbound direction carries more passengers in the afternoon and evening.
Figure 4-2. Passenger Demand Distribution
900
--- Inbound
800
-- Outbound
700
0
600
CL
WU
in 500
.S
400
(A
(A
am
300
200
100
0
5:15
AM
6:15
AM
7:15
AM
8:15
AM
9:15
AM
10:15 11:15 12:15
AM
AM
PM
1:15
PM
2:15
PM
3:15
PM
4:15
PM
5:15
PM
6:15
PM
7:15
PM
8:15
PM
9:15
PM
10:15 11:15 12:15
PM
PM
AM
Time
The vehicle fleet is composed mostly of low-floor, 60-foot, articulated buses that carry
approximately 100 passengers, while smaller 40-foot buses are typically used in the late
evening. The MBTA Silver Line buses are also low-emission Compressed Natural Gas
(CNG) vehicles, which are more environmentally-friendly. The buses have an
Automated Stop Announcement System (ASAS) and easy-access door ramp to comply
with the American with Disabilities Act (ADA). The MBTA Silver Line buses are
equipped with an Automatic Vehicle Location (AVL) system that uses Global Positioning
System (GPS) technology that transmits vehicle location data to the MBTA's Control
Center for real-time tracking and operations control. Also, buses have on-board sensors
to monitor fuel levels, engine conditions, odometer reading and other vehicle condition
indicators.
8 Passenger
boardings were obtained from ride checks recorded as part of a separate study on the Silver
Line on Washington Street route.
104
Case Study: Application to MBTA
4.2 AUTOMATED DATA COLLECTION SYSTEMS
The AVL system, provided by Siemens Transportation Systems, captures vehicle
location data both through interval polling and time-point based technology. Using GPS
technology, the system polls the location of each vehicle at time intervals to provide
updated tracking data. This data is integrated with a Geographic Information System
(GIS) that enables supervisors at the MBTA's Control Center to view the location data
through a graphical user interface. The system also provides access to the current
timetable and operator assignments which allows the interface to display whether a bus
is running early, on-time or late, as well as the operator ID. Time-at-location information
is also captured as buses traverse through a number of time points in the route. An
arrival-at-point timestamp is recorded as buses enter a zone around a specific set of
GPS coordinates that correspond to a time point in the route. The actual departure
timestamp is recorded as the buses leave this zone. The system is not equipped to
capture dwell time information such as the time stopped at a time point with the doors
open.
Polled data is transmitted over the air to the control center, while time-at-location data is
recorded in the on-board system, which is then downloaded at the garage at the end of
the day and archived in a central database. Thus, the polled data is mainly used for
real-time tracking, while the time-at-location data is used for off-line analysis.
The buses are not equipped with an Automatic Passenger Counter (APC) system. The
analysis process and service reliability evaluations are still applied to the Silver Line
route using the available AVL and integrated operator assignment data, but the lack of
passenger loads limits the ability to conduct detailed analyses in determining the causes
of unreliability on this route.
The system also displays real-time schedule information to passengers at each of the
stations, which are equipped with variable message signs. The on-board computer
system also has the capacity to capture information on fuel levels, odometer readings,
wheelchair ramp usage, engine conditions and alarm triggers.
4.2.1 ARCHIVED
DATA
The Automatic Vehicle Location (AVL) system installed in the MBTA Silver Line is
configured with 10 time point locations in both the inbound and outbound directions, and
at the Southampton Street garage. This means that almost all of the bus stops are
considered time points in the system (only the Lenox Street and Worcester Square bus
stops in both directions are not considered time points). Table 4-4 lists the 20 time
points and the corresponding identification numbers. Note that time point identifiers are
not unique for each direction, and both the Direction ID and Time Point ID must be used
to properly identify the location of a data record.
105
Chapter 4
Table 4-4. Silver Line Route Time Point Information
Stop Name
Southampton garage
Time Point
Sequence
Direction ID
Time Point ID
24
--
Inbound
Dudley Square
1
3
1
Melnea Cass Blvd.
2
3
6
Massachusetts Ave.
3
3
4
East Newton St.
4
3
12
Union Park St.
5
3
11
East Berkeley St.
6
3
5
Herald St.
7
3
9
New England Medical Center
8
3
7
Chinatown
9
3
8
Temple St.
10
3
3
Temple St.
11
4
3
Boylston St.
12
4
14
New England Medical Center
13
4
7
Herald St.
14
4
9
East Berkeley St.
15
4
5
Union Park St.
16
4
11
West Newton St.
17
4
26
Massachusetts Ave.
18
4
4
Melnea Cass Blvd.
19
4
6
Dudley Square
20
4
1
Outbound
Data recorded by the on-board system is downloaded into the MBTA's databases at the
end of service each night. The raw data is automatically pre-processed and sorted by
the system, and archived into several database tables. This research uses the most
comprehensive time-at-location data table, called "Time Point Crossing", where each
row is a distinct data record of a scheduled bus trip crossing a specific time point
location. Each data record contains values of the data items, listed in Table 4-5, that
characterizes the route, time point location, scheduled and actual times at the time point,
and vehicle and trip assignment.
106
Case Study: Application to MBTA
Table 4-5. Data Items of Time Point Crossing Record*
Data Item
Descriptiofn
Notes on Special Format
Unique ID
Unique idenliller to characterize data record
Numeric value
Calendar [D
Date of data record
Route ID
Route identifier
lyyymmdd (y-year, rn-month, d-day)
Specific assigned numeric value
Route Direction ID
Value to indicate inbound or outbound
direction
Specific assigned numeric value. For Silver Line
Washington St: 3 - Inbound, 4 - Outbound
Pattern ID
Combined with route ID, it identifies specific
route traveled
Specific assigned numeric value. For Silver Line
Washington St, there are only two pattem Ds that
Geographic location identifier
Numeric count of stops made so far per
Assigned numeric value specific to each time point
Cumulative numeric value. For Silver Line Washington
assignments
St. for example, a bus' first trip will start at 1 for the
garage, 2 for Dudley 3 for Metnea Cass, and 5 for
Mass. Ave. time point (4 is skipped because the Lenox
identify inbound and outbound travel.
Geo Node ID
Stop Offset
St. stop is not a tirme point
Scheduled Time
Scheduled time at imrne point
Numeric value corresponding to time in seconds past
midnight of day of service. Service spans from approx.
5:15am-1:15am.
Trips past midnight are same day
service, where time values are greater than 86400.
Actual Arrival Time
Aciual arrival time at time point
Same as Scheduled Time
Actual Departure Time
Daily Work Piece ID
Actual departure time from time point
Run assignment identifier
Same as Scheduled Time
Assigned numeric value
Time Point ID
identifier associated with specific time point
Assigned numeric value (described in Table 4-1)
Vehicle ID
Idenifier associated With bus operated
Assigned numeric value
Trip ID
Identifier associated with scheduled trip
Assigned numeric value
Pullout IlD
Identifier associated with specific block
Assidned numeric value
* There are six other data items in the table which do not contain any relevant data
(PLANNEDADHWAIVE R_IND, ADHERENCEWAIVER, WAIVER_ID, UPDATE_TIMESTAMP)
inconsistentdata (ODOMETER, IsRevenue)
or contain
Operator assignments for each scheduled trip are not included in the "Time Point
Crossing" table. Information on operators was found to be limited in the MBTA's
automated data collection system databases and difficult to reliably integrate with the
rest of the time point crossing location data9. However, operator assignments were
recovered from operator log sheets maintained by supervisors at the garage and
terminal and were manually entered for integration with the time point crossing data.
These log sheets contain information on daily work assignments, including the day,
departure time from garage or relief point (if work assignment begins at a point in the
route), run number (or work assignment id), bus number (differs from vehicle identifier in
107
9At the time of this research, a definition of the data items of each table and the relationship between
them was not available. This made it difficult to reliably interpret the bus operator data.
Chapter 4
system), badge number, and operator's full name. The log sheets also indicate whether
the trip was covered by the scheduled operator or whether a re-assignment occurred.
As previously mentioned, the on-board computer system also records information on
wheelchair ramp usage and other on-board sensors and alarm systems. However, the
data contained in these databases have rarely been used by the MBTA staff. This
research considers this data to be unreliable due to the lack of calibration tests to ensure
the archived data is accurate. While it would have been useful to determine whether a
schedule or headway deviation was caused by the use of a wheelchair ramp, this
research does not integrate these data items into the analysis process.
4.3
SCOPE OF ANALYSIS AND DATA ASSUMPTIONS
For this case study, the analysis process is applied to the Silver Line on Washington
Street on weekdays for the period of September 13 to October 7, 2004, considering
service throughout the entire day (5:15am to 1:15am). The first block of the analysis, the
characterization of service reliability, is also applied for the period of May 2 to May 27,
2005 since some important changes had occurred in the fareboxes over this period
which may have affected service reliability.
Daily time periods were those used by the MBTA and are as shown in Table 4-6.
Categorization of scheduled trips by time period is based on the trip's departure time
from the Dudley Square terminal.
Table 4-6. Silver Line Time Periods
Scheduled Headways
Tim e
Period
_
___
(minutes)
5:00 AM
5:45 AM
15
2
5:45 AM
6:40 AM
6-10
3
6:40 AM
9:05 AM
3-5
4
9:05 AM
1:00 PM
6-8
5
1:00 PM
2:00 PM
6-7
6
2:00:PM
3:55 PM
5- 6
7
3:55 PM
6:05 PM
3-5
8
6:05 PM
6:35 PM
5-6
9
6:35 PM
7:55 PM
10
10
7:55 PM
1:30 AM
12
1
For performance evaluation and analysis, three locations in each direction are
considered: the Dudley Square terminal, the Temple Place terminal, and East Berkeley
Street station. This allows for service reliability to be evaluated at the beginning, middle
and end of each trip. East Berkeley Street in considered a good mid-route time point
because it is near the transition point between the exclusive lane segment and the
downtown area, and previous ride checks shows this to be the maximum load point
on the Silver Line.
108
Case Study: Application to MBTA
Time point crossing data was obtained for all 19 weekdays in the first study period and
20 weekdays in the second study period, although not all scheduled trips contained valid
data. Data on actual trips may be missing due either to the trip not being made or the
AVL system failing to capture the location data. Capture rates varied, but overall capture
rates for the entire study period ranged between 70% and 95% of the scheduled trips
per time period. These capture rates are lower for headway evaluations because two
valid data records are needed to calculate the actual headway between consecutive
buses.
Operator assignments were recovered from paper log sheets obtained from the MBTA
staff for all days in the first study period except Monday, September 20, 2004. No
operator assignments were obtained for the May 2005 study period. The operator
assignments were manually entered into a database table and minor adjustments to the
data were made to integrate it with the time point crossing data. Integration was based
on departure time at the garage or scheduled departure time from the relief point in the
route, and the "Daily Work Piece ID" value in the data records. It was found that some of
the departure times were the same between the two sets of data and others were off by
5 minutes. Where scheduled times did not match, the first data record of each unique
"Daily Work Piece ID" data item served to identify the first trip of an operator's
assignment.
Assumptions also had to be made on operator assignments. Substitutions were
assumed for any log sheet record where the name of the scheduled operator was
scratched out and/or signed by a different operator. Records without a signature could
not be assumed to be missed service since in many cases actual time point crossing
data did exist. For these cases, where recorded data could not be matched with a
known operator assignment, the trips were marked with an unknown operator. Operator
type (part-time or full-time) was determined by the "Run Number ID" in the paper log
sheets, where runs operated by part-time operators have an ID number starting with 9.
4.4
SILVER LINE RELIABILITY ASSESSMENT
The proposed reliability analysis process discussed in Chapter 3 is applied to the Silver
Line on Washington Street route to evaluate its service reliability. Service reliability on
the Silver Line on Washington Street route is also evaluated using AVL data from May
2005. The results of both analyses are presented in this section.
4.4.1 RUNNING TIMES
At the trip level, scheduled and actual running times are compared on two segments of
the route: Dudley Square - East Berkeley Street, and East Berkeley Street - Temple
Place. The running time distributions for the two segments in the inbound and outbound
directions are calculated and summarized in Table 4-7 through Table 4-10. The tables
present the capture rate (number of trips with valid data divided by the number of
109
Chapter 4
scheduled trips), the scheduled and actual running time mean and standard deviations
of both, for all time periods.
Table 4-7. Running Time Analysis: Dudley to E. Berkeley - Inbound
Time Period
-# of Trips
# of Trips
Capture
Running Time (mins)
Mean
Start Time
End Time
Sched.
Obsv.
Rate
Sched.
5:00 AM
5:45 AM
38
33
86.8%
5:45 AM
6:40 AM
6:40 AM
9:05 AM
116
664
106
579
9:05 AM
1:00 PM
2:00 PM
1:00 PM
2:00 PM
3:55 PM
614
195
445
3:55
6:05
6:35
7:55
6:05 PM
6:35 PM
7:55 PM
1:30 AM
568
76
152
476
PM
PM
PM
PM
Std. Deviation
8.0
Obsv.
6.2
Sched.
0.0
Obsv.
1.0
91.4%
87.2%
8.0
10.8
7.6
10.0
0.0
1.5
1.5
2.2
529
154
369
86.2%
79.0%
82.9%
12.0
11.9
12.0
9.8
9.5
10.6
0.1
0.6
0.4
1.8
2.0
2.5
464
81.7%
94.7%
90.8%
70.0%
11.9
10.1
10.0
10.0
7.8
9.3
8.2
6.4
0.5
0.0
0.0
1.3
2.2
1.5
1.7
1.6
72
138
333
Table 4-8. Running Time Analysis: E. Berkeley to Temple - Inbound
Time Period
Start Time
5:00 AM
5:45 AM
6:40 AM
9:05 AM
1:00 PM
2:00 PM
3:55 PM
6:05 PM
End Time
5:45
6:40
9:05
1:00
2:00
3:55
6:05
6:35
AM
AM
AM
PM
PM
PM
PM
PM
Running Time (mins)
#of Trips
# of Trips
Capture
Sched.
Obvs.
Rate
38
116
664
614
195
445
568
76
33
106
579
518
152
368
460
72
86.8%
91.4%
87.2%
84.4%
77.9%
82.7%
81.0%
94.7%
4.0
4.0
5.0
5.0
5.0
5.0
5.0
4.2
4.9
6.1
7.9
7.0
7.0
7.5
5.0
90.8%
5.0
69.1%
4.3
6:35 PM
7:55 PM
152
138
7:55 PM
1:30 AM
476
329
Std
Sched.
sv.
Obsv.
6.6
0.0
0.0
0.2
0.0
0.1
0.1
0.0
0.0
0.6
0.9
1.2
2.0
1.6
1.6
2.1
1.3
6.1
0.0
1.6
5.0
0.4
1.6
Sched.Me n Obsv.
Table 4-9. Running Time Analysis: Temple to E. Berkeley - Outbound
Time Period
Trips
# of Trips
Capture
Running Time (mins)
End Time
# of Trp
Sched.
5:00 AM
5:45 AM
5:45 AM
6:40 AM
38
116
34
108
89.5%
93.1%
6.0
6.0
6:40 AM
9:05 AM
664
592
89.2%
6.1
9:05 AM
1:00 PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
7:55 PM
1:00 PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
7:55 PM
1:30 AM
614
195
445
568
76
152
476
510
167
398
474
70
145
330
83.1%
85.6%
89.4%
83.5%
92.1%
95.4%
69.3%
7.0
7.9
10.0
9.4
7.0
7.0
5.4
St
110
Tim
# of Ts
Obsv.
Cate
Rate
Mean
Sched.
Obsv.
Std. Deviation
Sched.
Obsv.
0.0
0.0
1.0
1.3
7.7
0.3
1.3
8.3
8.5
9.5
9.2
8.2
7.9
6.9
0.0
1.3
0.4
1.2
0.0
0.0
0.9
1.4
1.4
1.6
1.6
1.5
1.6
1.5
5.4
6.9
Case Study: Application to MBTA
Table 4-10. Running Time Analysis: E. Berkeley to Dudley - Outbound
Time Period
Start Time
End Time
#of Trips
Sched.
5:00 AM
5:45 AM
6:40 AM
5:45 AM
6:40 AM
9:05 AM
38
116
664
34
108
577
89.5%
93.1%
86.9%
8.0
8.0
8.8
5.6
6.5
7.3
0.0
0.0
1.1
2.0
1.8
1.9
9:05 AM
1:00 PM
614
477
77.7%
11.0
8.5
0.1
2.6
1:00 PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
7:55 PM
195
445
568
76
152
167
396
448
69
122
85.6%
89.0%
78.9%
90.8%
80.3%
10.4
10.0
9.7
9.0
9.0
8.0
8.6
8.6
7.7
7.1
0.7
0.2
0.4
0.0
0.0
2.0
1.7
2.2
1.3
1.5
7:55 PM
1:30 AM
476
309
64.9%
8.2
5.9
0.4
1.6
# of Trips
Obsv.
Capture
Rate
Meunning Timed. Diation
Obsv.
Sched.
Obsv.
Sched.
The running time analysis tables show that observed running time means are lower than
scheduled for the Dudley Square to East Berkeley Street segment in both the inbound
and outbound direction. For the East Berkeley Street to Temple Place segment, the
observed running time means are higher than scheduled in both directions. Standard
deviations vary across the route and throughout the day, with a tendency to be slightly
higher for the Dudley Square - East Berkeley Street segment
The running time distributions for the PM peak hour for the two segments in the both
directions are presented in Figure 4-3 through Figure 4-6. The AM peak hour graphs are
shown in Appendix A.
111
Chapter 4
Figure 4-3. PM Peak Running Time Distribution: Dudley to E. Berkeley - Inbound
500
450Scheduled
450
* Actual
Mean (Standard Deviation) of Running Time:
400
Scheduled: 11.9 (0.5) mins
Actual: 10.1 (2.2) mins
3501
300
250
0
200
150
100
50
0
5-7
7-9
9-11
11-13
13-15
15-17
17-19
19-21
21-23
Scheduled
0
0
33
431
0
0
0
0
0
Actual
16
131
197
86
14
14
3
1
2
Running Time (minutes)
Figure 4-4. PM Peak Running Time Distribution: E. Berkeley to Temple - Inbound
500
M Scheduled
450
400350
Mean (Standard Deviation) of Running Time:
Scheduled: 5.0 (0) mins
Actual: 7.5 (2.1) mins
-
300
(A
0.
f-
250
200
150
100 50
0
1-3
3-5
5-7
7-9
9-11
11-13
13-15
15-17
17-19
19-21
0
0
0
0
0
17
5
5
1
1
Scheduled
0
460
0
0
0
Actual
0
24
191
166
50
Running Time (minutes)
112
Case Study: Application to MBTA
Figure 4-5. PM Peak Running Time Distribution: Temple to E. Berkeley - Outbound
400-
Mean (Standard Deviation) of Running Time:
Scheduled: 9.4 (1.2) mins
Actual: 9.2 (1.6) mins
350
13 Scheduled
U Actual
300-
250U)
200-
0.
I-
150-
100-
50 1
0Scheduled
0
83
15
9-11
376
11-13
0
0
Actual
2
40
183
195
48
6
5-7
Running Time (minutes)
Figure 4-6. PM Peak Running Time Distribution: E. Berkeley to Dudley - Outbound
350 ,
0 Scheduled
(A
300
-
250
1
EActual
Mean (Standard Deviation) of Running Time:
Scheduled: 9.7 (0.4) mins
Actual: 8.7 (2.2) mins
200
0.
0
150
100
50
0
Scheduled1
Actual
5-7
7-9
9-11
11-13
13-15
15-17
17-19
19-21
0
119
329
0
0
0
0
0
72
230
116
16
2
6
1
1
11
21-23
0
4
Running Time (minutes)
113
Chapter 4
The running time graphs for the PM peak hour show how the observed running time
distributions for the Dudley Square - East Berkeley Street segment in both directions
have an observed mean running time lower than scheduled, but a higher variability. For
the East Berkeley Street - Temple Place segment, the observed mean and standard
deviation of running time are significantly higher than scheduled in the inbound direction,
and approximately equal in the outbound direction.
4.4.2 HEADWAY ADHERENCE
For headway analysis, the schedule and actual preceding headway of each bus as it
crosses a timepoint are calculated. Table 4-11 and Table 4-12 show the mean and
coefficient of variation of both scheduled and observed headways for Dudley Square,
East Berkeley Street and Temple Place in both directions for each time period.
Scheduled headway coefficients of variation are non-zero because headways vary by 1
or 2 minutes in each time period.
Table 4-11. Mean Scheduled and Actual Headways
Time Period
Sched.
Inbound
Actual (mins)
Outbound
Dudley
E. Berkeley
Temple
Temple
E. Berkeley
Dudley
14.9
10.1
4.2
7.2
5.9
4.9
4.4
7.0
15.9
10.3
4.3
7.3
6.0
5.0
4.2
6.8
16.1
10.4
4.3
7.2
5.6
5.0
4.2
6.9
14.1
10.4
4.3
7.2
5.6
5.0
4.2
7.2
14.3
10.7
4.2
7.2
5.7
5.0
4.1
7.0
13.6
10.9
4.2
7.1
5.6
4.9
4.2
6.4
Start Time
End Time
5:00 AM
5:45 AM
6:40 AM
9:05 AM
1:00 PM
2:00 PM
3:55 PM
6:05 PM
5:45 AM
6:40 AM
9:05 AM
1:00 PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
6:35 PM
7:55 PM
9.9
9.6
9.5
9.3
9.2
9.3
9.8
7:55 PM
1:30 AM
11.7
11.9
11.9
12.1
12.3
12.3
11.7
15.0
10.1
4.2
7.2
6.1
4.9
4.3
7.0
Table 4-12. Coefficients of Variation of Headways
Time Period
Inbound
Actual
Outbound
Dudley
E. Berkeley
Temple
Temple
E. Berkeley
Dudley
0.00
0.21
0.23
0.24
0.31
0.46
0.27
0.29
0.63
0.24
0.32
0.68
0.13
0.31
0.64
0.18
0.37
0.76
0.38
0.44
0.81
1:00 PM
2:00 PM
3:55 PM
0.12
0.08
0.08
0.41
0.39
0.50
0.62
0.54
0.70
0.68
0.60
0.75
0.66
0.53
0.76
0.73
0.66
0.87
0.82
0.79
0.98
6:05 PM
6:35 PM
7:55 PM
1:30 AM
0.13
0.14
0.03
0.07
0.57
0.37
0.33
0.18
0.72
0.47
0.41
0.24
0.78
0.55
0.47
0.27
0.77
0.51
0.44
0.33
0.86
0.55
0.54
0.35
0.95
0.64
0.58
0.35
Start Time
End Time
5:00 AM
5:45 AM
6:40 AM
5:45 AM
6:40 AM
9:05 AM
9:05 AM
1:00 PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
7:55 PM
114
Actual (mins)
Sched.
Case Study: Application to MBTA
Table 4-11 shows that mean observed headways are very close to scheduled values,
with the largest deviation between the mean scheduled and observed headway
occurring in the early AM peak period (5:15am-5:45am). However, the important
observation (from Table 4-12) is the actual coefficients of variation are high for all time
periods. This indicates headway variability is high, with buses tending to bunch together
and creating large gaps in service.
Variability increases as the buses traverse the route, with the highest values at the end
of the outbound trips. The exception, again, is at Temple Place, where a slight decrease
is typically observed in the headway variability as vehicles begin their outbound trip.
Thus, despite the heavily constrained nature of the Temple Place terminal, some
headway regulation and correction is being achieved there during the day.
Variability is highest in the PM peak period, suggesting that headway adherence
problems seem to propagate throughout the day until after the PM peak period, where
many buses return to the garage allowing headway performance to improve. Buses are
likely to be able to maintain better headways due to the decrease in demand, frequency
and traffic in the evening.
Figure 4-7 illustrates this variability by showing the headway ratio distribution for the PM
peak period.
Figure 4-7. PM Peak Headway Ratio Distribution
25%-Dudley
- Inbound
E. Berkeley - Inbound
Temple - Inbound
-
20%
-Temple
- Outbound
- E. Berkeley - Outbound
Dudley - Outbound
0 1 r5%1
/
0
0%
/
5%/
0%
<= 0.2
0.2-0.4
0.4-0.6
0.6-0.8
0.8-1
1-1.2
1.2-1.4
1.4-1.6
1.6-1.8
1.8-2
> 2
Headway Ratio
115
Chapter 4
Figure 4-7 shows that most trips leave Dudley Square with a headway ratio between 0.8
and 1.2, and the distribution at this point has a relatively low variability. At other points
on the route, the distributions show the variability of headway ratios, indicating poor
headway adherence. The percent of trips with headway ratios between 0.8 andl.2 is
less than 20% at all points except Dudley Square inbound. Trips are observed to have
a tendency for short headways (0.4 or less headway ratio) and very large headways (2.0
or more headway ratio).
4.4.3 PASSENGER WAIT TIMES
For high-frequency routes, the mean passenger wait time analysis is related to the mean
headway and its coefficient of variation. Expected passenger wait times characterize
reliability from the passengers' perspective, where the variability in headways affects the
mean expected passenger wait time. The expected passenger wait time (w) and
excess passenger wait time (EWT) equations, described in Section 3.6.1, are applied to
all trips in each time period and for each time point, with the results shown in Table 4-13
and Table 4-14, respectively.
Table 4-13. Expected Passenger Wait Times
Time Period
Start
End
Sched. Expect.
Pass. Wait Time
(mins)_________
Time
Time
Dudley
Inbound (mins)
E. Berkeley
Temple
Outbound (mins)
Temple
E. Berkeley
___
Dudley
_______
5:45 AM
6:40 AM
9:05 AM
1:00 PM
2:00 PM
6:40
9:05
1:00
2:00
3:55
AM
AM
PM
PM
PM
5.3
2.2
3.7
3.1
2.5
5.5
2.6
4.2
3.4
3.1
5.6
3.0
5.0
3.9
3.7
5.8
3.1
5.3
3.8
3.9
5.7
3.0
5.2
3.6
3.9
6.1
3.3
5.5
4.1
4.4
6.5
3.5
5.9
4.5
4.8
3:55 PM
6:05 PM
2.2
2.9
3.2
3.4
3.3
3.6
3.9
6:05 PM
6:35 PM
6:35 PM
7:55 PM
3.6
5.0
4.0
5.3
4.2
5.5
4.5
5.7
4.5
5.5
4.5
6.0
4.5
6.6
7:55 PM
1:30 AM
5.9
6.2
6.3
6.5
6.8
6.9
6.6
Table 4-14. Expected Excess Passenger Wait Times
Outbound (mins)
Inbound (mins)
Time Period
Start Time
End Time
Dudley
E. Berkeley
Temple
Temple
E. Berkeley
Dudley
5:45 AM
6:40 AM
6:40 AM
9:05 AM
0.3
0.3
0.3
0.8
0.5
0.9
0.4
0.8
0.8
1.1
1.2
1.3
9:05 AM
1:00 PM
0.6
1.3
1.6
1.5
1.9
2.3
PM
PM
PM
PM
PM
PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
7:55 PM
1:30 AM
0.4
0.6
0.7
0.4
0.4
0.3
0.8
1.3
1.0
0.6
0.6
0.4
0.8
1.4
1.2
0.9
0.7
0.6
0.5
1.4
1.1
1.0
0.5
0.9
1.0
1.9
1.4
1.0
1.0
1.0
1.5
2.4
1.8
0.9
1.6
0.7
1:00
2:00
3:55
6:05
6:35
7:55
These tables show deterioration in service as the buses traverse the route, with some
improvements at Temple Place before beginning the outbound trip. Service reliability
116
Case Study: Application to MBTA
continues to decrease in the outbound direction, with the highest value of excess
passenger wait times at Dudley Square (Outbound). Of course, no passengers actually
board at Dudley in the outbound direction since it is the end of the route, so these
numbers are theoretical only.
Table 4-14 shows passengers in the outbound direction will tend to wait, on average,
more than expected because headway variability at these points is higher. This will
have a greater impact in the PM peak when passenger demand is higher in the
outbound direction. Passengers in the midday period experience greater expected
excess wait times, particularly right after the AM peak and before the PM peak. The
ratio of excess wait time to scheduled expected passenger wait time is the highest for
these time periods. This shows that passengers in this period experience very high wait
times relative to that already expected (2 minutes of excess is significant if the
scheduled expected wait time is 3 minutes). On the other hand, excess wait times are
the lowest in the early morning and late night periods, where frequency is lower. This
means that reliability from the passengers' view is good at these time periods as they
should not expect to wait a lot more than scheduled (1 minute of excess wait time is not
as significant if the scheduled expected wait time is 6 minutes).
4.4.4 CHANGES IN PERFORMANCE:
MAY 2005
In January 2005, the MBTA installed a new Automatic Fare Collection (AFC) system on
the Silver Line on Washington Street route as an initial test of the technology as part of
deployment on the entire transit system. The AFC system created delays in the
boarding process due to problems with the new equipment, which was designed to
accept the new smartcard tickets, regular magnetic stripe cards, bills and coins.
However, a lack of familiarity with the new equipment combined with its greater
complexity and several design flaws contributed to an increase in dwell times and travel
times for the route.
In response to these problems, the MBTA increased scheduled running times on the
Silver Line on Washington Street corridor for certain time periods and segments of the
route. The May 2005 data reflect such changes in scheduled times and the reliability
analyses are intended to evaluate performance and assess any changes in performance
due to the implementation of the new AFC system.
Mean scheduled running times have been increased from the September-October period
in both segments in the inbound direction between 6:40am and 6:35pm (AM Peak,
Midday and PM peak periods). In the outbound direction, the scheduled running times
for in the East Berkeley Street - Temple Place segment have been increased for all time
periods except in the evening and late night (6:35pm-1:30am), and are approximately
the same for the Dudley Square - East Berkeley Street segment for all time periods.
Mean running times tend to be slightly higher for this analysis period (May 2005)
compared to those in the previous analysis period (September-October), except for the
early AM time period where mean actual running times have decreased.
117
Chapter 4
With regard to the standard deviations of running times, they are significantly higher for
the Dudley Square - East Berkeley Street inbound segment, especially in the PM peak
period. Because the majority of boardings occur on the first half of the route, the new
AFC equipment has impacted this segment of the route most negatively. Running time
variability on the East Berkeley Street - Temple Place inbound segment did not change
substantially, but variability has increased in the outbound direction.
The headway adherence results show that mean headways are close to scheduled,
but the coefficients of variation reveal headways are highly variable and unreliable
Variability tends to be worse in the outbound direction than in the inbound direction.
Compared to the September-October period, headway variability has increased in May
at most points and time periods. A decrease in headway variability is shown in the
early AM and in the outbound direction for the PM peak (and the period right before the
PM peak). This observation is unexpected, since variability should have increased for
this period and direction because a high percent of boardings in the outbound direction
occur in this segment of the route, and one would have expected the AFC equipment to
negatively impact this segment. The headway ratio distribution for the PM peak is
shown in Figure 4-8.
Figure 4-8. PM Peak Headway Distribution(May 2005)
40%
35%
0
30%
-Dudley - Inbound
E. Berkeley - Inbound
Temple - Inbound
-Temple - Outbound
-
E. Berkeley - Outbound
Dudley - Outbound
25%
20%
15%
10%
5%
0% i
<=
0 .4
i
0.4 to 0.8
i
0.8 to 1.2
Headway Ratio
118
i
i
1.2 to 1.6
1.6 to 2
>2
Case Study: Application to MBTA
Figure 4-8 shows that headway variability in the PM peak is high at all points on the
route. However, comparing this figure with Figure 4-7, it is observed that the headway
ratio distribution for Temple Place in the outbound direction is less variable and has a
more noticeable peak around better headway adherence.
As for expected passenger wait times, they are higher for the time periods between the
AM peak and PM peak in the May 2005 analysis period compared to the scheduled
values in the September-October period. Expected passenger wait times were observed
to increase for most of the points, with the exception again of the PM peak in the
outbound direction. The inbound direction in the early AM period is now shown to
experience lower-than-scheduled expected passenger wait times.
Comparing the expected excess passenger wait times from September-October and
May reveal an general increases, but a decrease in the inbound direction during the
early AM and late night periods, and again, in the outbound direction in the PM peak.
These results are consistent with those from the headway analyses, where a decrease
in headway variability in the PM peak outbound reflects an improvement in expected
passenger wait times for the same period and segment of the route.
4.4.5 OVERALL RELIABILITY ASSESSMENT
The characterization of service reliability of the Silver Line for both periods reveals the
following:
-
Running times between Dudley Square and East Berkeley Street are lower than
scheduled, which indicate lower mean travel times for passengers in this segment.
However, mean travel time between East Berkeley Street and Temple Place is
higher than scheduled. This indicates that vehicles tend to slightly recover from late
departures from Dudley Square in the inbound direction and improve on schedule
performance upon arrival back at Dudley Square.
-
The running times (and schedule deviations) standard deviations indicate a
problem for passengers who are unable to adjust their departure times (decision on
when to leave for a trip based on expected total travel time, including wait time and
in-vehicle time) because travel times (and departures) are unreliable and
inconsistent.
-
Mean observed headways are close to schedule. However, the coefficient of
variation shows that many buses operate with much lower or much higher headways,
creating reliability problems for both operations and passengers. Headway variability
increases as buses traverse the route (variability is higher in the outbound direction)
and over the day until after the PM peak period (coefficient of variation is the highest
highest in this period).
119
Chapter 4
-
The PM peak period is the worst in relation to operations (running times, schedule
deviations and headway adherence). The midday period is the worst in terms of
passenger wait times.
Comparing service reliability in May 2005 with September-October 2004, the following
observations are made:
4.5
-
Mean observed running times are higher in the May 2005, reflecting the
aforementioned operational problems with the new Automated Fare Collection
(AFC) system.
-
Headway variability seems to have worsened for most of the route and day,
compared to the September-October period, with more buses tending to bunch.
-
Expected passenger wait times have also increased for all time periods except the
early AM peak. Excess expected passenger wait times continue to be the worst in
the early Midday period (9:05am-1:00pm). However, decreases in the excess wait
time were observed in the early AM and late night (inbound direction) and the PM
peak period (outbound direction).
-
Reliability, as shown by the coefficient of variation of headways and passenger
wait times, has actually improved in the outbound direction in the PM peak, and thus
the scheduling adjustments appear to have had the desired effect for this period.
-
Overall results show that service reliability has worsened since the SeptemberOctober period, except for the outbound PM peak. Despite the increases in
scheduled running times, running time variability has increased significantly,
especially in the first half of the route (Dudley Square - East Berkeley
Street). In addition to the increase in travel times, variability in schedule deviations
and headway adherence has increased. This means that passengers experience
higher expected passenger wait times and poorer service quality.
IDENTIFYING THE CAUSES OF UNRELIABILITY
The second block of the proposed reliability analysis is applied to the Silver Line
Washington Street route to identify the causes of service unreliability. The results of
these analyses using the September-October 2004 data are presented in this section.
4.5.1 PERFORMANCE AT TERMINAL
The first step of the analysis is to evaluate performance at the terminals. There are two
terminals on the Silver Line Washington Street route: Dudley Square and Temple Place.
As previously noted (see Section 4.1), Temple Place is a physically constrained terminal
point and recovery times at this point are limited. Thus, only the results of the analysis
of the Dudley Square terminal are presented.
120
Case Study: Application to MBTA
As characterized in Section 4.4.2, the headway ratios at the Dudley Square terminal in
the inbound direction have a mean value of roughly 1.0 and a coefficient of variation that
varies between 0.18 and 0.57. The headway ratio distribution at the terminal is analyzed
to examine the departure behavior at the beginning of trips. Figure 4-9 illustrates the
number of trips that depart Dudley Square within a certain headway ratio range.
Figure 4-9. Headway Ratio Distributionat Dudley - Inbound
1400
1324
1200
1000
U)
800
0
600
415
400 1
315
200 1
193
132
73
0< 0.4
0.4 to
0.8
1.2 to 1.6
0.8 to 1.2
Headway Ratio
1.6 to 2
>2
Figure 4-9 shows that most trips (54%) departed Dudley Square for their inbound
journey within +/-20% of their scheduled headway. Extreme headway ratios (less
than 0.4 and more than 2.0) were observed for around 10% of the trips.
While the departure behavior at Dudley Square is better than at Temple Place, there are
still 46% of trips that are departing this terminal with short or long headways. This
departure behavior has an effect on service reliability on the rest of the route, especially
given the high trip frequency.
The next step in the analysis is to evaluate the effects of headway deviations at the
terminal on other points along the route. This is to examine the headway adherence
given the departure behavior from the terminal.
Trips are grouped by initial headway ratio at the Dudley Square terminal, and the mean,
standard deviation and distribution of the headway ratio at East Berkeley Street and
Temple Place are calculated for trips in each category. Because it cannot be
determined whether missing data is due to a missed trip or errors in data capture, the
analysis only uses records where valid headway data at the time point and the terminal
121
Chapter 4
(i.e., when there is data on two consecutive buses at that time point are known) are
available. The results of this analysis are summarized in Table 4-15.
Table 4-15. Headway Ratio at Inbound Time Points by Ratio at Dudley Terminal
Headway
Ratio at
Dudley
Dudley Square
Std.
Coeff.
Mean
Of
Square
East Berkeley
Std
Coeff.
Mean
D .
Of
Var.
Temple Place
Std
Coeff.
Mean
Dev
Of
Var.
Var.
0 to 0.4
0.4 to 0.8
0.8 to 1.2
1.2 to 1.6
1.6 to 2
0.23
0.64
1.00
1.35
1.80
0.12
0.11
0.10
0.11
0.11
0.51
0.17
0.10
0.08
0.06
0.38
0.55
1.02
1.33
1.61
0.38
0.37
0.42
0.53
0.57
1.02
0.67
0.41
0.40
0.36
0.38
0.58
1.03
1.31
1.61
0.30
0.40
0.52
0.58
0.70
0.80
0.69
0.50
0.44
0.44
>2
2.38
0.40
0.17
2.15
0.64
0.30
2.14
0.84
0.39
The table shows that, as expected, mean headway ratios remain approximately the
same along the route. The standard deviation shows that headway variability is higher
for trips with greater departure headways at Dudley Square. However, relative to the
mean, this variability is the highest for trips with initial headway ratios between
0 and 0.4 at Dudley Square.
The headway ratio distributions at East Berkeley in the inbound direction with respect to
the headway ratio at the Dudley Square are shown in Figure 4-10 and Figure 4-11.
The figures illustrate the distribution in terms of number of trips and probability of
headway ratio at this point given an initial headway ratio at Dudley Square. The data
table for Figure 4-10 and the distributions at Temple Place are shown in Appendix B. It
is noted that for this analysis, overtaking and possible missed trips are not included in
the analysis. Data records were included only where valid headway values (where
actual data was recorded for two consecutive trips) existed, and there was the same
preceding vehicle between the terminal and the time point in question.
122
Case Study: Application to MBTA
Figure 4-10. Headway Ratio Distribution at East Berkeley by Initial Ratio at Dudley
600
Initial Headway Ratio at
Dudley Square Terminal
/
->
500
+---
2
--
1.6 to 2
-
---
400
-----------------
-
1.6 to 1.2
-
0.8 to 1.2
-0.4toO.8
-
-<0.4
-
-
0
0.
I-W
300 +
200
/
+
-4.
100
-
0.
< 0.4
0;
P-
1.2 to 1.6
0.8 to 1.2
Headway Ratio at East Berkeley
0.4 to 0.8
1.6 to 2
>2
Figure 4-11. Headway Ratio Probability at East Berkeley by Initial Ratio at Dudley
0.80
Initial Headway Ratio at Dudley Square
>2
0.70 --1.6
- 0.8 to 1.2
- 0.4 to 0.8
-
0.60-
to 2
1.6 to 1.2
-<0.4
\
0.50
/
Probability of Headway Adherence
0.47
0.40-
0.30
0
0.20
00.0
< 0.4
0.4 to
0.8
0.8 to 1.2
1.2 to 1.6
1.6 to 2
> 2
Headway Ratio at East Berkeley
123
Chapter 4
As previously discussed, and as shown in Figure 4-10, about 54% of trips depart from
Dudley Square terminal with an initial headway within 20% of the scheduled headway.
Figure 4-10 also shows that most of these trips continue to have good headways at
East Berkeley Street. Trips with lower-than-scheduled headways at Dudley Square
tend to remain closely spaced, as shown by the distribution curve that is higher towards
low headway ratio values, and the trips with initial higher-than-scheduled headways
seem to continue with higher headways.
Figure 4-11 shows the expected behavior of headways at a point in the route given an
initial headway deviation at the terminal in terms of probabilities. Trips that depart with
an actual headway close to schedule tend to maintain a reasonable headway throughout
the route, with variability likely due to externalities and passenger demand. Trips
departing the terminal with a large preceding gap or bunched with its leader have a high
probability of increasing this gap or remaining bunched.
An important observation made from Figure 4-11 (and the other distributions shown in
Appendix B) is that the probability of actual headways being close to scheduled at other
points in the route decreases by almost half if the headway deviation at the terminal is
higher than 1.2 or lower than 0.8. This leads to the selection of these values as
threshold values for what is considered good headway adherence.
A similar analysis of schedule deviations showed that probability of buses arriving close
to the scheduled time at other points in the route varies along the route and by terminal
departure behavior, but the results show that buses departing the terminal between 1
minute early and 3 minutes late tend to be on-schedule at other points on the route.
Thus, these values are selected as the threshold values for what is considered good ontime performance.
The evaluation of performance at the terminal reveals the following:
-
Approximately half of the inbound trips depart the Dudley Square terminal with an
actual headway that is between 0.8 and 1.2 of their scheduled headway. A little less
than half of these trips are likely to arrive at East Berkeley Street at approximately
the scheduled headways (0.8-1.2 of the scheduled headway). This probability
decreases to about 40% at Temple Place. Deviations from scheduled headways at
the terminal are clearly a significant cause of unreliability.
-
These observations indicate that there are other important causes that are triggering
reliability problems, besides terminal deviations. As only about half of the trips that
depart the terminals without major deviations in headway are able to maintain
acceptable headway adherence, one must examine the data to determine other
causes along the route.
The causes of this high incidence of deviations at the terminals are evaluated by
applying the analyses described in Chapter 3. The process looks at schedules and
recovery times, operator and terminal supervisor behavior, heavy passenger loads and
other externalities. Limitations on available data from the MBTA, such as the lack of
124
Case Study: Application to MBTA
passenger counts and operator data from an automated data collection system,
constrain the level of detail in the analysis. Because data on passenger loads, terminal
supervision (and any control actions taken or applied) and traffic conditions are not
available, it is not easy to determine whether these elements are causes of unreliability
on this route. These possible causes are evaluated based on certain assumptions and
observations of the route, and not available AVL/APC data.
4.5.2 CAUSES OF DEVIATIONS AT TERMINALS
The analyses of available recovery times at terminals can help assess whether
scheduled times affect the departure behavior of trips from the terminal. If scheduled
recovery times are not sufficient to cover excess travel times of most trips, then buses
may not be able to begin their next trips following the scheduled headway.
Table 4-16 shows the mean scheduled, available and actual recovery times at Dudley
Square by time period. Available recovery time is the time between the arrival of a bus
at the terminal and the scheduled departure time of its subsequent trip. Actual recovery
time is the time between the arrival of a bus at the terminal and the actual departure time
of its next trip. The relative deviation in recovery is the difference between the mean
available and scheduled recovery times divided by the mean scheduled recovery time.
This represents the percent of deviation in recovery time, where a positive value means
buses had, on average, more than the scheduled recovery time, while a negative value
indicates they had less. For example, mean available recovery time was 28% lower
than the mean scheduled recovery time in both the AM peak and evening periods. The
mean observed schedule deviation is included as a reference to illustrate the mean
delay in trip departures.
Table 4-16. Mean Recovery Times at Dudley Square
Time Period
Trips
Mean Recovery Time (mins)
Relative
Deviation in
Mean Observed
Schedule
Start
End
Observed
Scheduled
Available
Actual
Recovery
Deviation (mins)
5:45 AM
6:40 AM
9:05 AM
1:00 PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
7:55 PM
6:40 AM
9:05 AM
1:00 PM
2:00 PM
3:55 PM
6:05 PM
6:35 PM
7:55 PM
1:30 AM
53
444
341
112
252
379
47
121
229
6.8
6.7
8.5
6.3
7.5
9.0
7.4
6.5
8.1
8.9
4.8
6.5
4.9
6.6
7.1
6.3
4.7
8.7
9.0
6.1
7.7
6.2
8.0
8.3
7.4
6.1
9.4
31%
-28%
-24%
-23%
-12%
-21%
-16%
-28%
8%
0.1
1.3
1.2
1.3
1.4
1.2
1.2
1.5
0.7
The table shows that at Dudley Square, in early morning and late night mean available
recovery times are higher than scheduled, and mean schedule deviations are the lowest.
Mean available recovery time is lower than scheduled for all other time periods, and
based on the relative deviation from the mean schedule recovery time, is the worst in the
AM peak and evening periods. For these time periods, the mean actual recovery time is
125
Chapter 4
higher than the mean available time, but close to the mean scheduled time (except in the
early midday). Trips in the evening time period also have the highest mean schedule
deviation from schedule. These trips are likely to be carrying over deviations from the
PM peak, and this is not accounted for in the scheduled recovery time.
The analysis of recovery times includes inferring the time buses typically spend boarding
and alighting passengers at the terminal. This is determined by estimating the time
buses with tight or negative available recovery time take at the terminal. Although
variability exists in the scheduled departure deviation, the analysis assumes that most
buses depart as soon as possible when pressured by scheduled times. The results
of this analysis are presented in Table 4-17, showing the number of trips within each
category.
Table 4-17. Available Recovery Time vs. Schedule Deviation at Dudley Square
Schedule Deviation
from Departure
-4 to -2 mins
Available Recovery Time
-2 to 0 mins
0 to 2 mins
2 to 4 mins
4 to 6 mins
3+ mins early
3-2 mins early
2-1 mins early
1-0 mins early
0-1 mins late
1-2 mins late
-------
-------
-----
0
0
4
5
0
---
-
-----
12
11
8
31
56
1
27
115
58
15
28
203
93
2-3 mins late
3-4 mins late
4-5 mins late
5-6 mins late
6-7 mins late
7-8 mins late
8-9 mins late
9-10 mins late
1
2
7
8
13
7
7
1
13
24
33
18
6
2
1
1
48
20
17
9
0
3
0
0
29
10
7
1
2
0
0
0
38
15
4
7
0
0
0
0
The distribution of schedule deviations given an available recovery time at Dudley
Square suggests that buses spend 2-3 minutes at the terminal to board and alight
passengers. This is inferred by observing that buses with negative available recovery
time (i.e., the bus arrived after the scheduled departure time of its next trip) tend to
leave 2-3 minutes later than their arrival, and buses with 0 to 2 minutes of recovery time
tend to depart 1 to 3 minutes late.
By determining that buses typically need 2-3 minutes for passengers to board and alight
at Dudley Square, then trips with available recovery times of less than 3 minutes do not
have enough time to make an on-time departure. The late departure in this case cannot
be attributed directly to operator behavior, and can be inferred to be the result of tight
schedules. Of course, these values will vary by time of day because of variability in
passenger demand, but a single value of 3 minutes is used because the analysis is
applied to all trips (not detailed by time of day or day of the week).
126
Case Study: Application to MBTA
The next step in the analysis is to infer the causes of unreliability by evaluating the
amount of available recovery time and the departure behavior from the terminal. As
described in Section 3.6.2:
-
Late departures of trips from the terminal that had no available recovery time are
caused by the late arrival of the previous trip.
-
Trips with available recovery times are further separated between those with very
short amount of recovery time, given by the threshold value determined as the time
buses typically need to board and alight passengers at the terminal (3 minutes for
Dudley Square), and those determined to have enough recovery time to account for
regular passenger demand and depart its next trip on-time.
-
For those trips which are determined to have enough recovery time, the actual
headway is examined, counting the trips with headway ratios greater than 1.2 and
those with a ratio lower than 0.8.
-
The schedule deviation in departure time is also considered, to determine whether
the bus had an on-time departure or not. This helps infer whether operators are
following schedules or any type of headway-control actions might have been
implemented by terminal supervisors. The trips are evaluated based on what the
actual headway would have been if the bus had departed on schedule, compared
to observed headway.
Figure 4-12 illustrates a flow diagram that presents the number (and percent) of trips that
result at each level of the sequential analysis. The percent value shown is in relation to
the number of trips in its parent category, and not the total number of trips.
127
Chapter 4
Figure 4-12. Analysis of Causes of Unreliabilityat Dudley Square
TwW iummbef
Irips ruoo
If
rmd
-1978
2M%of hiwpI vxe
Ift or no fc=4ty
Ii454 Consrdef
Tripswh
A~jlbIaieurowy
fime 0
17O (594)
Trips w
TUps wth
avalat16 rwmmyv
UvAiEbw recaliry
m4 0 to 3 mrirus
246 (12 4%)
/
S5% 17 a 7%)
A
or~ fVic' Hrihl
&ipa mdo
wo
wy
T ps taking
-
I1wh
rree
3 rmirtes at
tenmwal
Trips 1akig more
lian 3 mirutmts at
ls ImrrneI
T pevwith an
oervd hwvay
T&p Oith at
observd headway
iubo 0.8
mlio> 1.2
$10%
-
g4%)
r. hod v
-4.
(14J)a0A0
hcadw
!da"y lawo
Tri AhxIt 1alr
depariure ( 3
>
ninutes tals)
12
=43II18A%)
Tnpevw ti Ar ly
depatire ('I
inute eadlyl
= 21 16.0%)
Trkip aitii
wLa Tfpi %iMt an 1.dy|
departure ( I
deparnure (- 3
rnftnies lae)
= 11 (3K1
vrIlae efzy)
= 60 (16 8%)
and
xr upwris
-
Ptyshte headway-based m rt
:nd
schule daAaon rnMda
aPo1143
hea day bete r.
Poar supeMviskn aid eraar
bh~avar Qchdula disa2lon rta e
headway worse.
_M
Figure 4-12 shows that the majority of the trips have an available recovery time greater
than 3 minutes at Dudley Square. More than three-quarters of the inbound trips arrive from
their previous trip before the scheduled departure time, with "enough" recovery time (more
than 3 minutes) to account for normal passenger processing at the terminal and should depart
on schedule. From these results, it can be inferred that: a) scheduled times and
recovery times are not a significant cause of reliability problems at Dudley Square, as
almost 78% of the trips had enough time at the terminal before their scheduled departure
time; and b) there are other causes of reliability problems at the terminal because
128
Case Study: Application to MBTA
approximately 40% of these trips with available recovery time have poor headway
adherence (31% of the total trips) upon departure.
Evaluations of trips with adequate recovery time and poor headway adherence (both
less than and greater than scheduled headway) reveal that most do not have a schedule
deviation outside the "on-time performance" threshold values (1 minute early to 3
minutes late). This indicates that operators are following scheduled departure times, but
headway control is poor on the part of the terminal supervisor maintaining balanced
headways. For the off-schedule trips, headway-based control actions as a reason for the
departure deviations could explain only a few (32 of 620) trips where the deviation
made the actual headway better than it would have been if the trip had departed on-time.
But the deviation from schedule made the headway worse for a much greater number of
trips (103 out of 620). Operator behavior is then inferred to be the cause of this poor
departure performance.
For the other 21 % of the trips with little or no recovery time at Dudley Square, the
number of trips that spent more than 3 minutes at the terminal are evaluated. This
helps suggest whether operator behavior and poor terminal supervision are causing
reliability problems at this terminal by buses waiting at the terminal longer than normal.
The assumption is that trips, despite already being behind schedule, should depart the
terminal as soon as passenger boardings and alightings permit, in order to reduce the
magnitude of deviations. The threshold value to account for any passenger demand at
Dudley Square is 3 minutes. Any time spent after that is assumed to be caused by poor
operator behavior and poor terminal supervision.
The analysis shows that more than half of the trips (51.1%) with no available recovery
time spend more than 3 minutes at the terminal . For trips with 0 to 3 minutes of recovery
time, this percentage is 63%. The majority of these trips have poor headway adherence.
These observations lead to the conclusion that schedules (recovery times) are not a
major cause of poor performance at the terminal, but operator behavior and poor terminal
supervision are likely causes of reliability problems at Dudley Square.
However, this does not account for the effects that tight schedules might have on operator
behavior. Operators might consider that if available recovery times (also operator breaks)
are shorter than scheduled,, they have an added stress of keeping to tight schedules and
are not getting their deserved/scheduled break time, and will take this time, regarless of
the departure deviation. And while this behavior is controlled by the the operators, it may
still be a consequence of poor schedule planning.
While externalities and heavy passenger loads may also affect service performance at
the terminal, it is observed that the characteristics of Dudley Square station avert the
systematic incidence of these as major causes of unreliability at this terminal. Silver
Line buses have their own passenger bay at Dudley Square station, with two lanes and
capacity for 3-4 buses at any time. This means that buses can pass-up other
parked/idled buses, and are not affected by traffic conditions and private vehicles within
the terminal area. Even though Dudley Square is a major bus transfer station, schedule
129
Chapter 4
times are such that there is typically more than one Silver Line bus at the terminal, and
heavy passenger loads may be diverted to board the second bus to avoid overloading
one bus and causing major deviations in departure. The terminal supervisor is
responsible for this type of headway control and enforcement, instructing buses to
depart on schedule (or headway-based) and informing passengers to board the other
available buses.
A summary of the results of the analyses of Dudley Square reveals:
-
Recovery times at Dudley Square do not seem to be a significant cause of reliability
problems, as most trips have 3 or more minutes of available recovery time before
their inbound departure. Most of these trips are observed to have good headway
adherence and on-time performance. Poor departure performance appears to be
correlated with poor operator and terminal supervisor behavior, based on the number
of trips with high or low headways. This analysis does not account for the effects
that tight schedule planning might have on operator behavior with regard to
unofficial break times at the terminal.
-
Characteristics of Dudley Square station make it unlikely that externalities or heavy
passenger loads are systematic problems at this terminal. Accidents/breakdowns or
high passenger demand should not cause major departure time deviations at this
terminal, and should be handled by terminal supervisors, who can instruct buses to
overtake or depart despite passenger demand (instructing passengers to board the
next bus).
Operator Behavior
An analysis of operators is applied to trips at Dudley Square where operator behavior is
a potential cause of poor headway adherence. This analysis looks at the frequency of
poor performance of operators to help infer whether operator behavior is a systemic
problem. For trips with little or no recovery time and taking more than 3 minutes at the
terminal, only those with a large headway (greater than 1.2) are considered for this
operator analysis. Those with headway ratios lower than 0.8 are not counted because
inferring operator behavior is not easy, as the headway would have been smaller if the
trip had departed earlier (and not stayed at the terminal as long as it did).
Of the 206 trips with operator behavior as a possible cause of unreliability, operator
assignments (known to be covered by particular operator'0) were determined for 148 of
them. The number of poor performance trips is compared to the total number of trips
with valid headway data (actual headway is known) and the assigned operator, only 13
operators are observed to have performed poorly on more than 10% of their assigned
trips, as shown in Table 4-18.
Because operator assignments were obtained from supervisor log sheets from the garage, only duties
with a valid signature could be known to be covered by that particular operator.
10
130
Case Study: Application to MBTA
Table 4-18. Poor Performance Operators
Headway
headway data)
Trips
Recovered
Percent Of Poor
Trips (known
headway)
2595
5
15
15
100.0%
33.3%
2929
1
4
4
100.0%
25.0%
7483
22
90
121
74.4%
24.4%
7748
2
17
20
85.0%
11.8%
8407
13
54
90
60.0%
24.1%
9577
4
16
24
66.7%
25.0%
9703
13
95
138
68.8%
13.7%
65230
2
18
21
85.7%
11.1%
65376
1
1
8
12.5%
100.0%
65501
5
37
50
74.0%
13.5%
65541
3
16
24
66.7%
18.8%
12.5%
11.9%
Badge
# of Trips
with Poor
# of Assigned Trips
(with known
# of
Assigned
Percent Of
Assignments
65619
4
32
42
76.2%
67178
10
84
96
87.5%
It should be noted that this analysis only includes operator behavior as it affects departure
times at Dudley Square. Specifically, it does not take into account operator behavior
that affects running times. For example, it does not include operators that tend to run fast
(and bunched) to arrive early at the terminal and enjoy longer breaks (and still depart on
their next trip on-time). However, the results suggest the general tendencies in terms of
operator behavior, such as frequency of late trips, especially in the next step of the
analysis.
4.5.3 DEVIATIONS AT OTHER POINTS
The analysis now focuses on the causes of unreliability at other points on the route.
Previous sections noted that headway deviations at the terminals trigger reliability
problems on this route, but poor performance is still observed at points on the route even
when good headway adherence is observed at the terminal.
As described in Chapter 3, the process looks at deviations at the terminal, abnormal
passenger boardings and alightings, operator behavior and externalities. Reliability is
examined at East Berkeley, in both directions as a mid-route point, and headway
adherence is evaluated for headways within 20% of the scheduled headway.
Again, limitations on available data constrain the level of detail in the analysis. Inference
of abnormal passenger loads and traffic conditions as causes of unreliability on this
131
Chapter 4
route are not easy to determine and are evaluated based on certain assumptions and
observations (not available AVL/APC data). Results of the sequential steps to infer the
causes of deviations at East Berkeley Street are presented in the flow diagram shown in
Figure 4-13. The total number of trips reflects the number of trips with valid headway data
for both the terminal and East Berkeley Street.
Figure 4-13. Analysis of Causes of Unreliabilityat East Berkeley - Inbound
Total number of
trips analyzed
2353
Trips with poor headway
adherence
HR <0.8 849 (36.1%)
HR > 1.2
760 (32.3%)
Trips with poor terminal departure
behavior
At East Berkeley
At Dudley
HR> 1.2
HR < 0.8
S qua re
91
297
HR< 0.8
210
1
HR> 1.2
% .01 iE 2.S%
Total=
12 2
[began from
Dudley Square
with poor
headway
56.7% of trips with poor headway at
East Berkeley began trip with a poor
headway from Dudley Square
475 (295%) rFps W07 poxor
headway adherce had a hiogh
dweI tne at a stcp betwoen
Dudley and East Bikeley
High dweft time at time
points between Dtxdey
and East Borkaoy as
and indEcat Iof) of
abnormal passenger
L..
Abnormal passenger loads
inferred as a cause of
eviatwis at East Berkeley
Trips with high dwell time
_--100 seconds)
= 203 (292%)
toads
Operator Behavior
Externalities
to infer
Cannot be onferred
mdata.
from data
Figure 4-13 shows that 68.1% of the trips have poor headway adherence at East Berkeley,
and of these, nearly 57% had an initial headway deviation at the terminal. This indicates
that reliability is a big problem at East Berkeley Street station, and the main cause of these
is the poor headway adherence at the terminal that propagates and disrupts the headways
at other points in the route.
132
Case Study: Application to MBTA
For trips that departed the terminal within 20% of their scheduled headway, but still
showed poor headway adherence at East Berkeley, the results show that 29.2% of these
trips spent more than 100 seconds at a previous time point before arriving at East
Berkeley (time points in between include: Melnea Cass Blvd., Massachusetts Ave.,
Newton St. and Union Park St.). Because "door open" dwell times or passenger boarding
data is not recorded by the Silver Line's automated data collection systems, dwell times
are calculated indirectly using the actual arrival and departure time of vehicles at time
points. It is suspected that large dwell times are the result of abnormal passenger loads,
due to heavy passenger demand or the use of the wheelchair ramp, although in some
cases they may be caused by traffic signal delays.
Detailed analyses of the remaining 493 trips could not be applied to infer whether
operator behavior or externalities are a cause of poor headway adherence at this time
point.
Operator behavior is difficult to determine given the limited data available. However,
using the counts of trips with poor terminal departure that were inferred to be caused by
poor operator behavior (from the previous section) and the assignments of these 493
trips, a total number of possible off-schedule trips due to operator behavior are calculated
for each operator. The results of this analysis reveal that 44 operators (of 59 operators
found in the dataset) have more than 10% of assigned trips (with valid headway data)
within this set of possible deviated trips caused by operator behavior. A sample set of
these operators and resulting number of trips are shown in Table 4-19. While 44 out of
59 operators having poor headway performance is a significant percent, it is noted that
the capture rate of trips with known operator assignment and valid headway data is low
for a large number of these operators. This indicates that there is a large percent of an
operator's total assigned trips that are not included in this analysis due to missing AVL
data. Therefore, the analysis on operator behavior remains inconclusive and difficult
to interpret.
Table 4-19. Operator Behavior Analysis
Badge ID
# of Trips
with Poor
Headway
# of Assigned Trips
(with known
headway data)
# of
Assigned
Trips
Percent Of
Percent Of
Assignments
Determined
Poor Trips
(known
headway)
1943
19
88
222
39.6%
21.6%
2036
20
148
315
47.0%
13.5%
2112
23
112
216
51.9%
20.5%
2300
24
162
324
50.0%
14.8%
2595
5
15
15
100.0%
33.3%
35.3%
21.0%
2630
26
124
351
2703
11
80
150
53.3%
13.8%
2929
1
4
4
100.0%
25.0%
2957
16
130
204
63.7%
12.3%
133
Chapter 4
Externalities are also hard to infer as a cause of poor reliability on this route without
available data on speed profiles or reports on accidents or breakdowns. However, this
segment of the route is a semi-exclusive lane (bus-only and right-turn where permitted),
that is also at-grade and not protected (no barriers), which lessens the impacts of the
regular traffic stream but is still affected by illegal vehicle movements, double parking
and accidents.
A summary of the results of the analyses of causes of unreliability at East Berkeley
reveals:
4.6
-
Headway deviation at the terminals is the main cause of poor headway adherence
at East Berkeley. The analysis shows that both negative (lower than scheduled)
and positive (greater than scheduled) deviations of headways create variability in
headways at East Berkeley.
-
In the inbound direction, high dwell times at stops affect approximately 30% of the
trips that had good headways departing from Dudley Square but arrived at East
Berkeley with poor headway adherence. The lack of passenger load data,
wheelchair ramp usage and door open/close times, impedes the determination of
whether these high dwell times at time points between Dudley Square and East
Berkeley were due to abnormally high passenger loads, the use of the wheelchair
ramp, or long traffic signal delays.
-
Based on observations, aside from terminal deviations, it is believed that service
reliability problems on the inbound segment are caused by passenger loads, light
cycle delays, and operator behavior. Most boardings occur in the first half of the
routes, and the semi-exclusive lane allows operators to have a bit more control over
speeds (it also hinders the likelihood of traffic volumes affecting operations).
APPLICATION OF STRATEGIES
The following strategies are highlighted for application on this route, based on the results
of the previous analyses.
-
Terminal procedures. Terminal deviations were found to be the major cause of
unreliability on this route.
o At Dudley Square, trips tend to have more than 3 minutes of available
recovery time but have poor headway adherence because they are not
departing on-time or are remaining at the terminal for a long time. Thus,
good headway management by terminal supervisors is needed to maintain
regular headways (which were observed to be both large and small upon
arrival at Dudley Square). Terminal supervisors need to enforce better
operator behavior to avoid early or late departures that affect departure
headways and create problems on the route. Better terminal supervision
would also need to include enforcement to maintain the route clear of
134
Case Study: Application to MBTA
obstructions such as illegally parked vehicles or help buses depart the
station. Terminal supervisors should also be able to balance passenger
demand to avoid overcrowding and high dwell times. If supervisors know the
arrival of the next bus, they can instruct passengers to avoid loading one bus
and wait x minutes for the next one.
- Operator behavior, though it could only be directly inferred on a few trips in the
analysis, is still believed to cause service reliability problems, especially at the
terminals and in the semi-exclusive lane, where they have greater control of travel
speeds. Better operator training, along with improved supervision tactics to enforce
on-time performance and good headway adherence, are also noted as strategies
that would improve service reliability on this route.
- Corrective Strategies. Applied to both directions, corrective strategies would help
reduce the impacts of passenger demands and externalities on headway adherence
at points along the route, which were found to be the cause of poor headway
adherence at East Berkeley. Though more difficult in the mixed-traffic segment of the
route, the available AVL/CAD system provides the location of all buses and helps
target specific buses for corrective strategies to improve service reliability if traffic
conditions worsen or passenger demand increases.
- Traffic signal priority. Signal priority, especially in the mixed-traffic segment of the
route would improve service reliability and help maintain more balanced headways
by reducing the impacts of externalities such as heavy traffic volumes. Conditional
signal priority could also be applied to reduce the impact of operator behavior with
regard to running time variability (tendency of some operators to run early, drive
slower, etc.).
- Schedule planning. Although the results do not present much detail on this,
schedule adjustments are also recommended in order to better account for time of
day variability. Poor service reliability was observed in the time periods surrounding
the peak hours, which may be the result of inadequate schedule times that do not
account for possible peak demand or traffic conditions extending to these periods.
4.7
SUMMARY OF FINDINGS
The application of the proposed reliability framework described in Chapter 3 on the
Silver Line route was limited by the availability of data from the Silver Line Washington
Street route. The timepoint Automated Vehicle Location (AVL) data was useful in
assessing the service reliability on this route and make some inferences on the
causes of unreliability. However, data on passenger boardings, more detailed operator
assignments, travel speeds or accident/breakdown reports would have improved the
level of detail of this case study.
The assessments of service reliability of the Silver Line Washington Street corridor
reveals that although mean travel times and headways are close to schedule, variability is
quite significant on this route.
135
Chapter 4
Running times are lower than scheduled for the first half of the route (between Dudley
Square and East Berkeley Street), which allows buses to make up some of the
deviations with respect to schedules. However, variability in running times affects the
headways, leading to buses either becoming bunched or creating large gaps in service.
This variability, and overall service reliability, deteriorates along the second half of the
route between East Berkeley Street and Temple Place, where buses operate in mixed
traffic. Running times are higher than scheduled, and recovery times at Temple Place do
not give buses the opportunity to return to schedule. Thus, service unreliability continues
to deteriorate and is worse in the outbound direction, with high variability of service
attributes (running times, headways, expected wait times).
The results of the May 2005 Silver Line reliability assessment show further deterioration
of service reliability due to the introduction of the new Automatic Fare Collection (AFC)
system. Although scheduled running times were increased between September-October
2004 and May 2005, variability in running times and headways increased for this period,
creating greater problems in service reliability and increases in travel times and
expected wait times for passengers.
The main cause of service reliability problems are initial deviations from the terminal.
Buses are beginning their trip with poor headway adherence, which tends to propagate
and deteriorate service reliability as the buses traverse the route. Deviations at Dudley
Square are observed to be the result of poor operator behavior and terminal supervision.
Poor service reliability in the inbound direction is inferred to be caused by passenger
loads and operator behavior that deteriorates the headway of consecutive buses. In the
outbound direction, service reliability problems are inferred to be triggered by
externalities (traffic volumes, illegal vehicle movements, etc.) and passenger loads.
In terms of service operations, buses tend to be irregularly spaced (unbalanced
headways) and experience highly variable running times. Passengers seem to
experience unpredictable wait times and total travel times, and overcrowding, especially
in the peak hours where reliability deteriorates. For a Bus Rapid Transit (BRT) route,
characterized with features such as a semi-exclusive bus lane, high-capacity low-floor
vehicles, and Intelligent Transportation Systems (ITS) technologies, service reliability is
below the expected quality, and improvements such as better terminal headway controls
and corrective strategies are needed.
136
5.
CONCLUSIONS
This chapter summarizes the main elements of the proposed framework for analyzing
service reliability, the major findings from the application of the framework to the
Massachusetts Bay Transportation Authority's (MBTA) Silver Line Washington Street
route, and future extensions and research.
Section 5.1 briefly describes blocks of the proposed framework and key elements to
assess bus service reliability. The results of the case study on the Silver Line
Washington Route are presented in Section 5.2, along with a summary of
recommendations to the MBTA based on the major findings of this case study. Finally,
Section 5.3 suggests extensions to the proposed framework and areas of future
research.
5.1
FRAMEWORK SUMMARY
The proposed practical framework to assess service reliability explores the uses of
Automated Data Collection (ADC) systems characterize service reliability and evaluate
the causes of unreliability that may exist. It serves as a guide for transit agencies to
begin to analyze the large sets of data available from these systems to evaluate
performance and implement efficient strategies to improve service planning, operations
monitoring and management procedures.
A detailed review of the most significant causes of service reliability summarizes how
they impact service and outlines the complexities and interrelationship between different
causes. Also reviewed are the potential preventive and corrective strategies, and the
links between the causes of service unreliability and best strategy according to the
source of problems.
The proposed framework consists of three blocks: 1) characterizing service reliability
through service measures and performance reports; 2) identifying the causes of
reliability problems, starting with terminal deviations; and 3) selecting strategies which
target critical causes of unreliability to improve service.
Characterization of service reliability involves examining five key elements an agency
should analyze: a) available data inputs from ADC systems; b) output calculations from
data analysis; c) appropriate service measures; d) threshold values, and e) performance
reports.
Identification of causes includes two sequential processes to infer the causes of service
reliability problems. The first is focused on deviations at terminals because they serve
as control points to ensure on-time performance or headway adherence. Terminal
deviations also have a larger impact down the route as deviations tend to propagate.
The second process examines deviations at other points on the route, following a set of
137
Chapter 4
steps to infer the causes of unreliability: initial deviations at terminal, passenger loads,
poor schedule planning, operator behavior and externalities.
Application of strategies is based on the findings of the previous analysis, and includes
an assessment of the best strategies to prevent reliability problems and reduce the
impacts on service performance. This process also involves evaluating the practicality
of implementing such strategies, based on the cost of implementation and effectiveness
to improve service reliability on the route.
5.2
MAJOR FINDINGS
The application of the proposed framework to analyze service reliability on the Silver
Line Washington Street in Boston, MA served as the case study for this thesis.
The characterizations of service reliability reveals mean running times and mean
headways have a tendency to be close to schedule. However, the standard deviations
and coefficient of variations of the distributions show that variability of service attributes
is high. This indicates that bus arrivals and passenger wait times are unpredictable and
travel times are irregular. As a Bus Rapid Transit route, headway adherence is poor on
this route, with a tendency of buses to bunch together and leave gaps in service.
The case study also included an analysis of service reliability with recent data (May
2005) to evaluate the impacts of the implementation of a new Automatic Fare Collection
(AFC) system. The results of this analysis showed that mean running times and
headway variability increased from the previous assessment. This indicates that service
reliability has deteriorated, with increased travel times and expected passengers wait
times.
The main cause of service unreliability on this route was identified to be deviations at the
terminals. Trips are departing the terminal with poor headway adherence (and therefore,
poor on-time performance), which propagates and creates further reliability problems
down the route. At Dudley Square, a major bus transfer point with a dedicated loading
area for Silver Line buses, a combination of poor terminal supervision and operator
behavior are inferred to be cause of most of the deviated trips. Recovery times,
externalities and passenger loads at this terminal are inferred to cause only minor
problems. However, the analysis does not account the effects that scheduled times
might have on operator behavior. At other points in the route, operator behavior and
passenger loads are observed to affect reliability in the inbound direction.
As for strategies to improve service reliability, emphasis is given to better supervision at
the terminal. Terminal supervision is needed to enforce good operator behavior and
balance departure headways to prevent poor headway adherence from propagating
down the route. Supervisors at terminals are also able to apply control strategies, such
as holding or instructing buses to depart regardless of schedule times, to maintain
headways, and able to coordinate passenger loads to avoid overcrowding of buses.
138
Case Study: Application to MBTA
Along the route, preventive strategies such as operator training and corrective strategies
are highlighted as potential strategies to reduce the variability in running times and
balance headways to reduce the occurrence of bunches and gaps in service. Traffic
signal priority would also improve service reliability by reducing variability, especially in
the mixed-traffic portion of the route.
5.3
FUTURE RESEARCH
The case study application of the proposed framework for reliability assessment
revealed a number of areas for future extensions to the framework and bus service
reliability research.
Application of the framework on another route. The Silver Line Washington Street route
provided limited time-at-location and operator assignment data. The lack of passenger
loads (boardings and alightings), more detailed event records, such as doors open and
doors close, and speed profiles constrained the analysis on inferring the causes of
unreliability on this route. A case study of the proposed framework on a route which has
more detailed data available would prove useful and necessary to evaluate the practical
application of this process.
Follow-up reliability assessment of the Silver Line. The first block of the analysis was
applied to the Silver Line in May 2005 to characterize reliability after the implementation
of a new Automatic Fare Collection (AFC) system. The second part of the analysis, the
identification of causes, can be applied to this period to compare the results of inferred
causes of unreliability of this route and further evaluate the impacts of the AFC system
on service reliability.
Development of models. Empirical and simulation models ought to be developed to
further explore the interrelationship between causes of service unreliability. The
complexities of the most significant causes of service unreliability were not fully
analyzed. The proposed framework explored the most basic interrelationships to
provide a guide for transit providers to begin understanding the dynamics of service
reliability on a route. Regression models would provide a more detailed analysis
quantifying the impacts of each cause in a more simultaneous process, rather than the
sequential steps described. Simulations models could also be developed to better
integrate variations by time of day, operator behavior, traffic conditions and weather, into
the analysis of service reliability.
139
BIBLIOGRAPHY
Abkowitz, M, Slavin, H., Waksman, R., Englisher, L., Wilson, N. "Transit Service Reliability'.
Report UMTA-MA-06-0049-78-1. USDOT Transportation Systems Center. Cambridge, MA.
1978.
Abkowitz, M. "The Transit Service Reliability Problem and Potential Solutions. Proceedings
of the August 1982 Transit Reliability Workshop". Sponsored by USDOT Urban Mass
Transportation Administration. 1983.
Benn, H.P. and Barton-Aschman Associates, Inc. "Bus Route Evaluation Standards".
Transit Cooperative Research Program. Synthesis 10. Washington, D.C. 1995.
Bowman, L.A., Turnquist, M.A. "Service Freuqncy, Schedule Reliability and Passenger Wait
Times at Transit Stops". Transportation Research. Vol. 15A. No. 6. 1981.
Eberlein, X.J., Wilson, N.H.M., Barnhart, C., Bernstein, D. "The Real-Time Deadheading
Problem in Transit Operations Control," Transportation Research (Part B), Vol. 32, No. 2,
pp. 77-100. 1998.
Furth, P.G. "Data Analysis for Bus Planning and Monitoring". Synthesis of Transit Practice
34. Transportation Research Board. Washington, D.C. 2000.
Furth, P.G. and Muller, T.H.G. "Conditional Bus Priority at Signalized Intersections". Paper
presented at the 79th Annual Meeting of the Transportation Research Board. Washington,
DC. 2000.
Furth, P.G., Hemily, B.J., Muller, T.H.J., Strathman, J.G. "Uses of Archived AVL-APC Data
to Improve Transit Performance and Management: Review and Potentiaf'. TCRP Web
Document H-28. Transportation Research Board. Washington, D.C. 2003.
Hammerle, M. "Analysis of Automatic Vehicle Location and Passenger Count Data for the
Chicago Transit Authority'. Master's Thesis. University of Illinois at Chicago. 2004.
Kimpel, T., Strathman, J.G., Callas, S. "Improving Scheduling Through Performance
Monitoring Using A VL and APC Data". Portland, OR. 2004.
Kimpel, T., Strathman, J., Bertini, R., Callas, St. "Analysis of Transit Signal Priority Using
Archived TriMet Bus Dispatch System Data". Portland, OR. 2005.
Levinson, Herbert S. "Supervision Strategies for Improved Reliability of Bus Routes".
NCTRP Synthesis of Transit Practice 15, TRB, National Research Council, Washington, DC.
1991.
141
Appendix A
Levinson, H.S., Zimmerman, S., Clinger, J., Gast, J., Rutherford, S., Bruhn, E. "Bus Rapid
Transit. Volume 2: Implementation Guidelines". TCRP Report 90. Transportation Research
Board. Washington, D.C. 2003.
Moses, I.E. "The Value of Real-Time Information for Disruption Response in a Transit
System". Master's Thesis. Massachusetts Institute of Technology. 2005.
"Transit Capacity and Quality of Service Manual - 2 nd Edition". TCRP Project Report 100.
Transportation Research Board. Washington, D.C.
Turnquist, M.A. Strategies for Improving Reliability of Bus Transit Service. Transportation
Research Record 818. Pages 7-13. 1981.
Service Delivery Policy. Massachusetts Bay Transportation Authority. 1996.
Strathman, J.G., Dueker, K.J., Kimpel, T., Gerhart, R., Turner, K., Taylor, P., Callas, S.,
Griffin, D. "Service Reliability Impacts of Computer-Aided Dispatching and Automatic Vehicle
Location Technology: A Tri-Met Case Study'. Portland, OR. 1999.
Strathman, J.G., Kimpel, T., Dueker, K.J., Gerhart, R., Callas, S. "Evaluation of Transit
Operations: Data Applications of Tri-Met's Automated Bus Dispatch System". Portland, OR.
20011.
Strathman, J.G., Dueker, K., Kimpel, T., Gerhart, R., Turner, K., Callas, S., Griffin, D. "Bus
Transit Operations Control: Review and an Experiment Involving Tri-Met's Automated Bus
Dispatch System". Journal of Public Transportation, 41(1). 20012.
Strathman, J.G., Kimpel, T., Callas, S. "Headway Deviation Effects on Bus Passenger
Loads: Analysis of Tri-Met's Archived AVL-APC Data". Portland, OR. 2003.
Wile, Erik S. "Use of Automatically Collected Data to Improve Transit Line Performance".
Master's Thesis. Massachusetts Institute of Technology. 2003.
Wilson, N.H.M., Nelson, D., Palmere, A., Grayson, T.H., Cedequist, C. "Service-Quality
Monitoring for High-Frequency Transit Lines". Transportation Research Record 1349.
Pages 3-11. 1992.
Chicago Transit Authority website. www.transitchicaqo.orq
New York City Transit. Metropolitan Transportation Authority. www.mta.nvc.nv.us/nvct/
Massachusetts Bay Transportation Authority website. www.mbta.com (Silver Line route:
www.allaboutsilverline.com)
142
APPENDIX A: RUNNING TIME DISTRIBUTIONS
Running Time Distribution
Inbound: Dudley to East Berkeley - AM Peak
350
* Scheduled
300
- EActual
EActual-
300
250
Mean (Standard Deviation) of Running Time:
Scheduled: 10.8 (1.5) mins
Actual: 10.0 (2.3) mins
200
0.
I-
150
100
50
0
Scheduled
Actual
0-5
5-7
7-9
0
0
16
36
151
171
1
11-13
I
13-15
9-11
1
114
197
298
0
33
131
Running Time (minutes)
1
15-17
17-19
-19-21
21-23
0
5
0
3
0
1
0
2
Running Time Distribution
Inbound: East Berkeley to Temple - AM Peak
700
SMScheduled
600
Mean (Standard Deviation) of Running Time:
Scheduled: 5.0 (0.2) mins
Actual: 6.1 (1.2) mins
E Actual
500
U)
400
.
300
200
100
0
Scheduled
Actual
1-3
0
0
3-5
579
89
I
I
7-9
5-7
0
0
105
373
Running Time (minutes)
9-11
0
11
11-13
0
1
143
Appendix A
Running Time Distribution
Outbound - Temple Place to East Berkeley - AM Peak
700
600
GScheduled
EActual
Mean (Standard Deviation) of Running Time:
Scheduled: 6.1 (0.4) mins
Actual: 7.7 (1.3) mins
500
400
U)
0.
300
200
100
9-11
7-9'
5-7
Scheduled
0
592
0
0
0
Actual
12
159
332
84
5
Running Time (minutes)
Running Time Distribution
Outbound: East Berkeley to Dudley Square - AM Peak
500
1 Scheduled
450
N Actual
Mean (Standard Deviation) of Running Time:
Scheduled: 8.8 (1.1) mins
Actual: 7.3 (1.9) mins
400
350
300
U)
250
0.
I-
200
150
100
50
0
0-5
5-7
7-9
9-11
11-13
13-15
15-17
17-19
19-21
0
0
0
0
0
8
6
4
2
1
Scheduled
0
0
474
103
Actual
13
304
208
30
Running Time (minutes)
144
21-23
1
APPENDIX B: TERMINAL ANALYSIS
Headway Ratio at Dudley Square Terminal
0.4
0.8 to 1.2
0.4 to
to 0.8
0.8
1.6 to 2
1.6 to 1.2
Headway Ratio at
East Berkeley
>2
< 0.4
0.4 to 0.8
0.8 to 1.2
0
1
2
2
7
8
15
27
54
90
218
567
1.2 to 1.6
1.6 to2
>2
4
6
26
25
26
21
93
51
20
Total # of Trips
39
89
260
Temple Place
>2
1.6 to 2
1.6 to 1.2
< 0.4
1
4
< 0.4
Total #
Total #
135
125
55
65
20
4
307
398
690
242
71
28
15
4
1
4
0
1
383
158
97
1216
335
94
2033
Headway Ratio at Dudley Square Terminal
Headway Ratio at
0.8 to 1.2
0.4 to 0.8
< 0.4
Total #
18
121
136
58
338
98
63
24
8
411
591
21
3
1
322
0
1
0
91
333
158
133
1964
0.4 to 0.8
0.8 to 1.2
1
3
8
13
32
49
248
455
1.2 to 1.6
1.6 to2
>2
Total # of Trips
8
5
20
38
17
21
22
85
76
52
24
251
211
76
66
1177
Headway Ratio Distribution at Temple Place - Inbound
500
Initial Headway Ratio at
Dudley Square Terminal
450 +-
X ---------------
------------- /
400 4---350 - - - - - - - - -
--- ------------- -
300
/
2
-1.6
to 2
-1.2
N
to 1.6
- 0.8 to 1.2
-
0.4 to 0.8
-
U,
0.
I-
->
0.4
-Oto
-/
250
200
150
- - -- -- -- -100
-
-
-
---- -----------------
--
------
--- -- - -- -- -
- -- --
-
-
50-
0< 0.4
0.4 to 0.8
0.8 to 1.2
1.2 to 1.6
1.6 to 2
>2
Headway Ratio at Temple Place
145
Appendix B
Probability of Headway Ratio at Temple - Inbound
0.70
Initial Headway Ratio at Dudley Square
---
0.60-
>
-
1.2 to 1.6
0.8 to 1.2
-0.4 to 0.8
-
-
0.50
2
1.6 to 2
-
Oto
--
0.4
SProbability of Good Adherence
0.40
,
-25
0.39,
0..
000
0.20
0..20
0.100.15
0.00
< 0.4
146
.
0.4 to
0.8
1.2 to 1.6
0.8 to 1.2
Headway Ratio at Temple Place
1.6 to 2
> 2